diff --git a/client/src/components/BackgroundTasks/BackgroundTasksWidget.extra.test.tsx b/client/src/components/BackgroundTasks/BackgroundTasksWidget.extra.test.tsx
index b48fe3221..4ba36c96b 100644
--- a/client/src/components/BackgroundTasks/BackgroundTasksWidget.extra.test.tsx
+++ b/client/src/components/BackgroundTasks/BackgroundTasksWidget.extra.test.tsx
@@ -304,3 +304,44 @@ describe('BackgroundTasksWidget: the mark on a finished job (#2477)', () => {
expect(baseElement.querySelector('svg.lucide-alert-triangle')).toBeNull()
})
})
+
+describe('BackgroundTasksWidget — a receipt scanned from Costs', () => {
+ const RECEIPT = { merchant: 'Café', date: '2026-09-20', total: 12.5, currency: 'EUR', items: [] }
+
+ it('FE-W4BGT-025: says it is reading the receipt while the job runs', () => {
+ useBackgroundTasksStore.setState({ tasks: [task({ status: 'running', kind: 'costs', label: 'bill.jpg', items: undefined })] })
+ render(
)
+ expect(screen.getByText('Reading the receipt…')).toBeInTheDocument()
+ })
+
+ it('FE-W4BGT-026: import:done with a read receipt offers to review the expense, which opens the trip', () => {
+ useBackgroundTasksStore.setState({ tasks: [task({ status: 'running', kind: 'costs', label: 'bill.jpg', items: undefined })] })
+ render(
)
+ const handler = vi.mocked(addListener).mock.calls[0][0] as WsHandler
+
+ act(() => { handler({ type: 'import:done', jobId: 'j1', tripId: 't1', result: { receipt: RECEIPT, warnings: [] } }) })
+ expect(useBackgroundTasksStore.getState().tasks[0].receipt).toEqual(RECEIPT)
+
+ fireEvent.click(screen.getByRole('button', { name: 'Review expense' }))
+ expect(useBackgroundTasksStore.getState().tasks[0].reviewRequested).toBe(true)
+ expect(navigate).toHaveBeenCalledWith('/trips/t1')
+ })
+
+ it('FE-W4BGT-027: a photo nothing could be read from says so, with its warning and no AI retry', async () => {
+ vi.mocked(healthApi.features).mockResolvedValue({ bookingImport: true, aiParsing: true })
+ const photo = new File(['x'], 'bill.jpg', { type: 'image/jpeg' })
+ useBackgroundTasksStore.setState({ tasks: [task({ kind: 'costs', receipt: null, warnings: ['bill.jpg: no receipt could be read'], sourceFiles: [photo] })] })
+ render(
)
+ expect(screen.getByText('No receipt could be read from this photo.')).toBeInTheDocument()
+ expect(screen.getByText('bill.jpg: no receipt could be read')).toBeInTheDocument()
+ await waitFor(() => expect(healthApi.features).toHaveBeenCalled())
+ expect(screen.queryByRole('button', { name: 'Try AI parsing' })).not.toBeInTheDocument()
+ })
+
+ it('FE-W4BGT-028: the reload reconcile reads a receipt job\'s status too', async () => {
+ vi.mocked(reservationsApi.importJobStatus).mockResolvedValue({ status: 'done', done: 1, total: 1, result: { receipt: RECEIPT, warnings: [] } as never })
+ useBackgroundTasksStore.setState({ tasks: [task({ kind: 'costs', items: undefined })] })
+ render(
)
+ expect(await screen.findByRole('button', { name: 'Review expense' })).toBeInTheDocument()
+ })
+})
diff --git a/client/src/components/BackgroundTasks/BackgroundTasksWidget.tsx b/client/src/components/BackgroundTasks/BackgroundTasksWidget.tsx
index d6de54f34..6240e5e55 100644
--- a/client/src/components/BackgroundTasks/BackgroundTasksWidget.tsx
+++ b/client/src/components/BackgroundTasks/BackgroundTasksWidget.tsx
@@ -6,7 +6,18 @@ import { useTranslation } from '../../i18n'
import { addListener, removeListener } from '../../api/websocket'
import { reservationsApi, healthApi } from '../../api/client'
import { saveImportFiles } from '../../db/offlineDb'
-import { useBackgroundTasksStore, type BackgroundImportTask } from '../../store/backgroundTasksStore'
+import { useBackgroundTasksStore, taskFoundSomething, type BackgroundImportTask } from '../../store/backgroundTasksStore'
+import type { ReceiptRead } from '@trek/shared'
+
+/**
+ * One reading of a finished job's result, wherever it arrives from (the push,
+ * the reload reconcile, the poll): a booking parse answers `items`, a receipt
+ * scan `receipt`.
+ */
+function settleJob(id: string, tripId: string, result: unknown): void {
+ const r = result as { items?: unknown[]; warnings?: string[]; receipt?: ReceiptRead | null } | undefined
+ useBackgroundTasksStore.getState().setDone(id, tripId, (r?.items ?? []) as never, r?.warnings ?? [], r?.receipt)
+}
/**
* Global, route-independent widget (bottom-right) that tracks background booking
@@ -20,7 +31,6 @@ export default function BackgroundTasksWidget() {
const navigate = useNavigate()
const tasks = useBackgroundTasksStore((s) => s.tasks)
const setProgress = useBackgroundTasksStore((s) => s.setProgress)
- const setDone = useBackgroundTasksStore((s) => s.setDone)
const setError = useBackgroundTasksStore((s) => s.setError)
const requestReview = useBackgroundTasksStore((s) => s.requestReview)
const dismiss = useBackgroundTasksStore((s) => s.dismiss)
@@ -66,7 +76,7 @@ export default function BackgroundTasksWidget() {
reservationsApi
.importJobStatus(task.tripId, task.id)
.then((s) => {
- if (s.status === 'done') setDone(task.id, task.tripId, (s.result?.items ?? []) as never, s.result?.warnings ?? [])
+ if (s.status === 'done') settleJob(task.id, task.tripId, s.result)
else if (s.status === 'error') setError(task.id, task.tripId, s.error ?? 'error')
else setProgress(task.id, task.tripId, s.done, s.total)
})
@@ -87,14 +97,12 @@ export default function BackgroundTasksWidget() {
const tripId = String(e.tripId ?? '')
if (!id) return
if (type === 'import:progress') setProgress(id, tripId, Number(e.done ?? 0), Number(e.total ?? 1))
- else if (type === 'import:done') {
- const result = e.result as { items?: unknown[]; warnings?: string[] } | undefined
- setDone(id, tripId, (result?.items ?? []) as never, result?.warnings ?? [])
- } else if (type === 'import:error') setError(id, tripId, String(e.message ?? 'error'))
+ else if (type === 'import:done') settleJob(id, tripId, e.result)
+ else if (type === 'import:error') setError(id, tripId, String(e.message ?? 'error'))
}
addListener(handler)
return () => removeListener(handler)
- }, [setProgress, setDone, setError])
+ }, [setProgress, setError])
// Backstop: poll jobs whose state we still need — running ones (in case a WebSocket push
// was missed) and a restored 'done' task whose items haven't been re-fetched yet (so a
@@ -107,7 +115,7 @@ export default function BackgroundTasksWidget() {
reservationsApi
.importJobStatus(task.tripId, task.id)
.then((s) => {
- if (s.status === 'done') setDone(task.id, task.tripId, (s.result?.items ?? []) as never, s.result?.warnings ?? [])
+ if (s.status === 'done') settleJob(task.id, task.tripId, s.result)
else if (s.status === 'error') setError(task.id, task.tripId, s.error ?? 'error')
else setProgress(task.id, task.tripId, s.done, s.total)
})
@@ -120,7 +128,7 @@ export default function BackgroundTasksWidget() {
}
}, 5000)
return () => clearInterval(iv)
- }, [tasks, setProgress, setDone, setError, t])
+ }, [tasks, setProgress, setError, t])
if (tasks.length === 0) return null
@@ -141,11 +149,11 @@ export default function BackgroundTasksWidget() {
>
{(task.status === 'running' || (task.status === 'done' && task.items === undefined)) &&
}
- {task.status === 'done' && task.items !== undefined && task.items.length > 0 &&
}
+ {task.status === 'done' && task.items !== undefined && taskFoundSomething(task) &&
}
{/* A parse that found nothing is not a success, whatever the job
status says: the same card used to show a green tick next to
"no reservations could be extracted" (#2477). */}
- {task.status === 'done' && task.items?.length === 0 &&
}
+ {task.status === 'done' && task.items !== undefined && !taskFoundSomething(task) &&
}
{task.status === 'error' &&
}
@@ -156,7 +164,7 @@ export default function BackgroundTasksWidget() {
{task.status === 'running' && (
- {t('reservations.import.parsing')}
+ {t(task.kind === 'costs' ? 'costs.scan.reading' : 'reservations.import.parsing')}
{task.total > 1 ? ` · ${task.done}/${task.total}` : ''}
)}
@@ -165,14 +173,14 @@ export default function BackgroundTasksWidget() {
task.items === undefined ? (
// Restored from a reload; items are being re-fetched (see the poll backstop).
{t('reservations.import.parsing')}
- ) : task.items.length > 0 ? (
+ ) : taskFoundSomething(task) ? (
review(task)}
className="bg-accent text-accent-text"
style={{ marginTop: 4, border: 'none', borderRadius: 8, padding: '4px 12px', fontSize: 'calc(11.5px * var(--fs-scale-caption, 1))', fontWeight: 600, cursor: 'pointer', fontFamily: 'inherit' }}
>
- {t('common.import')}
+ {t(task.kind === 'costs' ? 'costs.scan.review' : 'common.import')}
{/* A partly-understood import warns too. Warnings used to be
shown only when nothing at all was found, so anything the
@@ -188,14 +196,14 @@ export default function BackgroundTasksWidget() {
) : (
- {t('reservations.import.previewEmpty')}
+ {t(task.kind === 'costs' ? 'costs.scan.nothingRead' : 'reservations.import.previewEmpty')}
{(task.warnings?.length ?? 0) > 0 && (
{task.warnings!.join('\n')}
)}
- {aiParsing && task.mode !== 'force-ai' && task.sourceFiles && task.sourceFiles.length > 0 && (
+ {aiParsing && task.kind !== 'costs' && task.mode !== 'force-ai' && task.sourceFiles && task.sourceFiles.length > 0 && (
retryWithAi(task)}
disabled={retrying === task.id}
diff --git a/client/src/components/Budget/CostsPanel.helpers.test.ts b/client/src/components/Budget/CostsPanel.helpers.test.ts
index bd655873f..6e1fc7586 100644
--- a/client/src/components/Budget/CostsPanel.helpers.test.ts
+++ b/client/src/components/Budget/CostsPanel.helpers.test.ts
@@ -1,5 +1,5 @@
import { describe, it, expect } from 'vitest'
-import { calculateTicketShares, finalBudgetFor, finalBudgetSources, hasTicketSplit, paidByUser, payerSum, payersBalanced, readTicketItems, readUserNote, rebalancePayers, settlementDate, splitCents, splitEqualShares, writeTicketItems, type TicketItem } from './CostsPanel.helpers'
+import { calculateTicketShares, expenseEditorFor, newExpenseSeed, receiptToPrefill, finalBudgetFor, finalBudgetSources, hasTicketSplit, paidByUser, payerSum, payersBalanced, readTicketItems, readUserNote, rebalancePayers, settlementDate, splitCents, splitEqualShares, writeTicketItems, type TicketItem } from './CostsPanel.helpers'
describe('splitCents', () => {
it('splits evenly when it divides cleanly', () => {
@@ -310,3 +310,55 @@ describe('finalBudgetSources', () => {
.toEqual({ fronted: [], moved: [], outstanding: [] })
})
})
+
+describe('newExpenseSeed', () => {
+ it('starts a plain new expense on the base currency, today, with nothing to attach', () => {
+ expect(newExpenseSeed(undefined, 'eur', [1, 2], '2026-09-24')).toEqual({
+ currency: 'EUR', day: '2026-09-24', total: '', ticketItems: [], receiptFiles: [],
+ })
+ })
+
+ it('reads a booking price in the base currency, padded to its decimals', () => {
+ expect(newExpenseSeed({ amount: 4.9 }, 'EUR', [], '2026-09-24').total).toBe('4.90')
+ })
+
+ it('seeds a scanned receipt: its currency, day, lines shared by everyone, and the photo', () => {
+ const photo = new File(['x'], 'bill.jpg')
+ const seed = newExpenseSeed(
+ { amount: 1500, currency: 'jpy', date: '2026-09-20', lines: [{ name: 'Ramen', price: 1200 }, { name: 'Tea', price: 300 }], receiptFiles: [photo] },
+ 'EUR', [1, 2], '2026-09-24',
+ )
+ expect(seed).toMatchObject({ currency: 'JPY', day: '2026-09-20', total: '1500', receiptFiles: [photo] })
+ expect(seed.ticketItems.map(i => [i.name, i.price, [...i.participants]])).toEqual([['Ramen', '1200', [1, 2]], ['Tea', '300', [1, 2]]])
+ })
+})
+
+describe('receiptToPrefill', () => {
+ it('turns a read receipt into an expense prefill, leaving out what was not read', () => {
+ const photo = new File(['x'], 'bill.jpg')
+ expect(receiptToPrefill({ merchant: 'Café', date: null, total: 12.5, currency: 'EUR', items: [] }, [photo])).toEqual({
+ name: 'Café', amount: 12.5, currency: 'EUR', date: undefined, lines: [], receiptFiles: [photo],
+ })
+ expect(receiptToPrefill({ merchant: null, date: '2026-09-01', total: null, currency: null, items: [] }, [])).toMatchObject({
+ name: undefined, amount: undefined, currency: undefined, date: '2026-09-01',
+ })
+ })
+})
+
+describe('expenseEditorFor', () => {
+ const closeBooking = () => {}
+ const closeReceipt = () => {}
+
+ it('is nothing when neither opener asks', () => {
+ expect(expenseEditorFor(null, closeBooking, null, closeReceipt)).toBeNull()
+ })
+
+ it('opens a scanned receipt as a new expense', () => {
+ expect(expenseEditorFor(null, closeBooking, { name: 'Café' }, closeReceipt)).toEqual({ key: 'receipt', editing: null, prefill: { name: 'Café' }, close: closeReceipt })
+ })
+
+ it('lets the booking win, since it is what was just clicked', () => {
+ expect(expenseEditorFor({ editing: null, prefill: { amount: 3 } }, closeBooking, { name: 'Café' }, closeReceipt))
+ .toEqual({ key: 'booking', editing: null, prefill: { amount: 3 }, close: closeBooking })
+ })
+})
diff --git a/client/src/components/Budget/CostsPanel.helpers.ts b/client/src/components/Budget/CostsPanel.helpers.ts
index 9aa6d401a..678571546 100644
--- a/client/src/components/Budget/CostsPanel.helpers.ts
+++ b/client/src/components/Budget/CostsPanel.helpers.ts
@@ -11,8 +11,10 @@
* Amounts are the raw input strings, parsed on use (same as customAmounts).
*/
-import type { BudgetParticipantFinal } from '@trek/shared'
-import { currencyDecimals } from '../../utils/formatters'
+import type { BudgetParticipantFinal, ReceiptRead } from '@trek/shared'
+import { amountToInputString, currencyDecimals } from '../../utils/formatters'
+import type { ExpensePrefill } from './CostsPanel'
+import type { BudgetItem } from '../../types'
// The split and receipt fields guard their own precision on every keystroke, so the
// guard has to follow the currency: a three-decimal one (KWD, BHD, …) seeds three
@@ -264,3 +266,59 @@ export function calculateTicketShares(items: TicketItem[]): { shares: Record ({
+ id: `receipt-${i}`,
+ name: line.name,
+ price: amountToInputString(line.price, currency),
+ participants: new Set(peopleIds),
+ })),
+ receiptFiles: prefill?.receiptFiles ?? [],
+ }
+}
+
+/** A scanned receipt as the prefill of a new expense, with its photo to attach on save. */
+export function receiptToPrefill(receipt: ReceiptRead, photos: File[]): ExpensePrefill {
+ return {
+ name: receipt.merchant ?? undefined,
+ amount: receipt.total ?? undefined,
+ currency: receipt.currency ?? undefined,
+ date: receipt.date ?? undefined,
+ lines: receipt.items,
+ receiptFiles: photos,
+ }
+}
+
+/**
+ * The expense editor a trip page shows, if any. Two things open it: a booking's
+ * Costs block (edit that expense, or start one from the booking's price) and a
+ * receipt scanned from Costs. The booking wins, since it is what the person just
+ * clicked; `key` remounts the editor when one hands over to the other, so each
+ * starts from its own seed.
+ */
+export function expenseEditorFor(
+ booking: { editing: BudgetItem | null; prefill?: ExpensePrefill } | null,
+ closeBooking: () => void,
+ receipt: ExpensePrefill | null,
+ closeReceipt: () => void,
+): { key: string; editing: BudgetItem | null; prefill?: ExpensePrefill; close: () => void } | null {
+ if (booking) return { key: 'booking', editing: booking.editing, prefill: booking.prefill, close: closeBooking }
+ if (receipt) return { key: 'receipt', editing: null, prefill: receipt, close: closeReceipt }
+ return null
+}
diff --git a/client/src/components/Budget/CostsPanel.test.tsx b/client/src/components/Budget/CostsPanel.test.tsx
index d970c160a..8003fe279 100644
--- a/client/src/components/Budget/CostsPanel.test.tsx
+++ b/client/src/components/Budget/CostsPanel.test.tsx
@@ -1625,6 +1625,33 @@ describe('CostsPanel — expense modal', () => {
expect(onSaved).toHaveBeenCalled()
})
+ it('FE-W5COSTS-081: offers Scan receipt beside Add expense only when the AI model reads images', async () => {
+ server.use(http.get('/api/llm/capabilities', () => HttpResponse.json({ images: true })))
+ const { unmount } = render( {}} onSaved={() => {}} />
+ )
+
+ expect(screen.getByDisplayValue('Sushi Dai')).toBeInTheDocument()
+ // Yen has no minor unit, so the amount is not padded.
+ expect(screen.getByDisplayValue('4200')).toBeInTheDocument()
+ expect(screen.getByText('bill.jpg')).toBeInTheDocument()
+ })
+
// The mobile sheet already dates a new expense by the traveller's own clock;
// the desktop modal filed it under the UTC day, so the same expense entered
// late in Tokyo or early in Los Angeles landed on a different day per surface.
diff --git a/client/src/components/Budget/CostsPanel.tsx b/client/src/components/Budget/CostsPanel.tsx
index 9bbf395a8..3ec72ca97 100644
--- a/client/src/components/Budget/CostsPanel.tsx
+++ b/client/src/components/Budget/CostsPanel.tsx
@@ -1,6 +1,6 @@
import { Fragment, useState, useEffect, useMemo, useCallback } from 'react'
import { useSearchParams } from 'react-router'
-import { ArrowDown, ArrowUp, BarChart3, Plus, Search, ArrowRight, ArrowLeftRight, Check, RotateCcw, Pencil, Trash2, AlertCircle, Download, StickyNote, ChevronDown, Receipt, Paperclip } from 'lucide-react'
+import { ArrowDown, ArrowUp, BarChart3, Plus, Search, ArrowRight, ArrowLeftRight, Check, RotateCcw, Pencil, Trash2, AlertCircle, Download, StickyNote, ChevronDown, Receipt, Paperclip, ScanLine } from 'lucide-react'
import { useTripStore } from '../../store/tripStore'
import { useAuthStore } from '../../store/authStore'
import { useSettingsStore } from '../../store/settingsStore'
@@ -19,11 +19,13 @@ import Modal from '../shared/Modal'
import CustomSelect from '../shared/CustomSelect'
import { CustomDatePicker } from '../shared/CustomDateTimePicker'
import { localToday } from '../Planner/today'
+import { useReceiptScan } from './useReceiptScan'
+import { ReceiptScanModal } from './ReceiptScanModal'
import { SYMBOLS, currenciesWith, SPLIT_COLORS } from './BudgetPanel.constants'
-import { amountPattern, calculateTicketShares, finalBudgetFor, finalBudgetSources, hasTicketSplit, NOTE_MAX, paidByUser, payersBalanced, readTicketItems, readUserNote, rebalancePayers, settlementDate, splitEqualShares, writeTicketItems, type TicketItem } from './CostsPanel.helpers'
+import { amountPattern, calculateTicketShares, finalBudgetFor, finalBudgetSources, hasTicketSplit, NOTE_MAX, paidByUser, payersBalanced, readTicketItems, newExpenseSeed, readUserNote, rebalancePayers, settlementDate, splitEqualShares, writeTicketItems, type TicketItem } from './CostsPanel.helpers'
import { COST_CATEGORY_LIST, catMeta } from './costsCategories'
import { ReceiptPreviewModal } from './ReceiptPreviewModal'
-import type { BudgetParticipantFinal, BudgetUnconverted } from '@trek/shared'
+import type { BudgetParticipantFinal, BudgetUnconverted, ReceiptLine } from '@trek/shared'
import type { BudgetItem, BudgetItemReceipt } from '../../types'
import type { TripMember } from './BudgetPanelMemberChips'
import GuestBadge from '../shared/GuestBadge'
@@ -83,6 +85,7 @@ export default function CostsPanel({ tripId, tripMembers = [] }: CostsPanelProps
const me = useAuthStore(s => s.user?.id ?? -1)
const can = useCanDo()
const canEdit = can('budget_edit', trip)
+ const receiptScan = useReceiptScan(tripId, canEdit)
const toast = useToast()
const { t, locale } = useTranslation()
const isMobile = useIsMobile()
@@ -387,7 +390,8 @@ export default function CostsPanel({ tripId, tripMembers = [] }: CostsPanelProps
0}
- onSettleAll={settleAll} onAddExpense={() => { setEditing(null); setModalOpen(true) }} />
+ onSettleAll={settleAll} onAddExpense={() => { setEditing(null); setModalOpen(true) }}
+ onScanReceipt={receiptScan.offered ? receiptScan.open : undefined} />
{/* ── Summary cards ── */}
@@ -509,6 +513,8 @@ export default function CostsPanel({ tripId, tripMembers = [] }: CostsPanelProps
)}
+ setModalOpen(false)}
@@ -634,6 +640,13 @@ export default function CostsPanel({ tripId, tripMembers = [] }: CostsPanelProps
)}
+ {receiptScan.offered && (
+
+
+ )}
{/* Owe / Owed */}
@@ -1237,6 +1250,11 @@ export interface ExpensePrefill {
reservationId?: number
/** Set when the expense is being created from a place (#1298). */
placeId?: number
+ /** The rest comes from a scanned receipt: its currency, day and lines, and the photo, attached on save. */
+ currency?: string
+ date?: string
+ lines?: ReceiptLine[]
+ receiptFiles?: File[]
}
export function ExpenseModal({ tripId, base, people, me, editing, prefill, onClose, onSaved }: {
@@ -1252,17 +1270,16 @@ export function ExpenseModal({ tripId, base, people, me, editing, prefill, onClo
const [name, setName] = useState(editing?.name || prefill?.name || '')
const [cat, setCat] = useState
(editing ? catMeta(editing.category).key : (prefill?.category || 'food'))
- const [currency, setCurrency] = useState(editingCurrency)
- const [day, setDay] = useState(editing?.expense_date || localToday())
+ const [seed] = useState(() => newExpenseSeed(prefill, base, people.map(p => p.id), localToday()))
+ const [currency, setCurrency] = useState(editing ? editingCurrency : seed.currency)
+ const [day, setDay] = useState(editing ? (editing.expense_date || localToday()) : seed.day)
const [note, setNote] = useState(() => readUserNote(editing))
// Edit and prefill seeds are padded to the currency's decimals (#2175): the DB
// returns numbers, so a saved 4,90 would otherwise reopen as "4,9" and a saved
- // 5,00 as "5". A prefill has no currency of its own — it is read as `base`,
- // which is also what the currency field starts on.
+ // 5,00 as "5".
const [total, setTotal] = useState(() => {
if (editing) return editing.total_price ? amountToInputString(editing.total_price, editingCurrency) : ''
- if (prefill?.amount != null) return amountToInputString(prefill.amount, base)
- return ''
+ return seed.total
})
const [participants, setParticipants] = useState>(() =>
editing ? new Set((editing.members || []).map(m => m.user_id)) : new Set(people.map(p => p.id)))
@@ -1303,7 +1320,7 @@ export function ExpenseModal({ tripId, base, people, me, editing, prefill, onClo
return 'equally'
})
- const [ticketItems, setTicketItems] = useState(() => readTicketItems(editing))
+ const [ticketItems, setTicketItems] = useState(() => editing ? readTicketItems(editing) : seed.ticketItems)
const [customAmounts, setCustomAmounts] = useState>(() => {
const m: Record = {}
@@ -1318,7 +1335,7 @@ export function ExpenseModal({ tripId, base, people, me, editing, prefill, onClo
})
const [receipts, setReceipts] = useState(() => editing?.receipts || [])
- const [pendingReceiptFiles, setPendingReceiptFiles] = useState([])
+ const [pendingReceiptFiles, setPendingReceiptFiles] = useState(() => editing ? [] : seed.receiptFiles)
const [uploadingReceipt, setUploadingReceipt] = useState(false)
const [modalPreviewReceipts, setModalPreviewReceipts] = useState<{ receipts: BudgetItemReceipt[]; initialIndex: number } | null>(null)
diff --git a/client/src/components/Budget/CostsToolbar.test.tsx b/client/src/components/Budget/CostsToolbar.test.tsx
index 3c7fbc9ad..7b1e3ada8 100644
--- a/client/src/components/Budget/CostsToolbar.test.tsx
+++ b/client/src/components/Budget/CostsToolbar.test.tsx
@@ -1,4 +1,4 @@
-// FE-COSTSBAR-001 to FE-COSTSBAR-007
+// FE-COSTSBAR-001 to FE-COSTSBAR-008
import { describe, it, expect, vi } from 'vitest'
import { render, screen, fireEvent } from '../../../tests/helpers/render'
import CostsToolbar from './CostsToolbar'
@@ -66,4 +66,18 @@ describe('CostsToolbar', () => {
expect(screen.queryByRole('button')).toBeNull()
expect(screen.getByText('8 days')).toBeInTheDocument()
})
+
+ it('FE-COSTSBAR-008: offers Scan receipt only when handed a way to open it', () => {
+ const onScanReceipt = vi.fn()
+ const { unmount } = render(
+ '#000'} canEdit canSettle
+ onSettleAll={vi.fn()} onAddExpense={vi.fn()} onScanReceipt={onScanReceipt} />,
+ )
+ fireEvent.click(screen.getByRole('button', { name: 'Scan receipt' }))
+ expect(onScanReceipt).toHaveBeenCalledTimes(1)
+ unmount()
+
+ setup()
+ expect(screen.queryByRole('button', { name: 'Scan receipt' })).toBeNull()
+ })
})
diff --git a/client/src/components/Budget/CostsToolbar.tsx b/client/src/components/Budget/CostsToolbar.tsx
index 97b034db4..ece9ae37b 100644
--- a/client/src/components/Budget/CostsToolbar.tsx
+++ b/client/src/components/Budget/CostsToolbar.tsx
@@ -1,4 +1,4 @@
-import { Check, Plus } from 'lucide-react'
+import { Check, Plus, ScanLine } from 'lucide-react'
import { useTranslation } from '../../i18n'
import type { TripMember } from './BudgetPanelMemberChips'
@@ -13,6 +13,8 @@ interface CostsToolbarProps {
canSettle: boolean
onSettleAll: () => void
onAddExpense: () => void
+ /** Opens the receipt scan; left out when the AI model reads no images. */
+ onScanReceipt?: () => void
}
const BODY_SIZE = 'calc(13px * var(--fs-scale-body, 1))'
@@ -22,7 +24,7 @@ const BODY_SIZE = 'calc(13px * var(--fs-scale-body, 1))'
* Bookings, Lists and Files open with: the tab's name, then what the numbers
* cover (the trip's days and who travels), and the two actions on the right.
*/
-export default function CostsToolbar({ dateMeta, people, me, colorFor, canEdit, canSettle, onSettleAll, onAddExpense }: CostsToolbarProps) {
+export default function CostsToolbar({ dateMeta, people, me, colorFor, canEdit, canSettle, onSettleAll, onAddExpense, onScanReceipt }: CostsToolbarProps) {
const { t } = useTranslation()
const chip = 'inline-flex items-center whitespace-nowrap rounded-full bg-surface-card px-3 py-1.5 font-medium text-content-muted shadow-sm'
const button = 'inline-flex items-center gap-1.5 rounded-[10px] border-0 px-[14px] py-[9px] font-medium hover:opacity-[0.88]'
@@ -64,6 +66,12 @@ export default function CostsToolbar({ dateMeta, people, me, colorFor, canEdit,
+
+ )}
Scan receipt }
+ (null)
+ const [dragOver, setDragOver] = useState(false)
+ if (!scan.isOpen) return null
+
+ const onDrop = (e: React.DragEvent) => {
+ e.preventDefault()
+ setDragOver(false)
+ scan.choose(Array.from(e.dataTransfer.files))
+ }
+
+ return createPortal(
+ { if (e.target === e.currentTarget) scan.close() }}>
+
+
+
{t('costs.scan.title')}
+
+
+
+
+
+
{t('costs.scan.accepted')}
+
{ scan.choose(e.target.files ? Array.from(e.target.files) : []); e.target.value = '' }} />
+
inputRef.current?.click()}
+ onDragOver={(e) => { e.preventDefault(); setDragOver(true) }}
+ onDragEnter={(e) => { e.preventDefault(); setDragOver(true) }}
+ onDragLeave={(e) => { if (e.target === e.currentTarget) setDragOver(false) }}
+ onDrop={onDrop}
+ className={`mb-3 flex min-h-[100px] w-full flex-col items-center justify-center gap-1.5 rounded-xl border-2 border-dashed p-4 text-body font-medium transition-colors ${dragOver ? 'border-accent bg-surface-tertiary' : 'border-edge bg-transparent'}`}>
+
+ {dragOver ? t('costs.scan.dropActive') : scan.photo ? scan.photo.name : t('costs.scan.dropHere')}
+
+
+ {scan.error && (
+
{scan.error}
+ )}
+
+
+
+
+ {t('common.cancel')}
+
+
+ {scan.starting ? t('costs.scan.reading') : t('costs.scan.start')}
+
+
+
+
(null)
+ const [error, setError] = useState('')
+ const [starting, setStarting] = useState(false)
+
+ useEffect(() => {
+ if (!canEdit) return
+ let live = true
+ llmRepo.readsPhotos().then((v) => { if (live) setReadsPhotos(v) }).catch(() => { if (live) setReadsPhotos(false) })
+ return () => { live = false }
+ }, [canEdit])
+
+ const open = () => { setPhoto(null); setError(''); setIsOpen(true) }
+ const close = () => { if (!starting) setIsOpen(false) }
+
+ /** Take the first of the picked or dropped files, if it is a photo the model can read. */
+ const choose = (files: File[]) => {
+ const picked = files[0]
+ if (!picked) return
+ const ext = '.' + (picked.name.toLowerCase().split('.').pop() ?? '')
+ if (!(LLM_PHOTO_EXTENSIONS as readonly string[]).includes(ext)) { setError(t('costs.scan.unsupported')); return }
+ if (picked.size > MAX_PHOTO_BYTES) { setError(t('reservations.import.fileTooLarge', { name: picked.name })); return }
+ setPhoto(picked)
+ setError('')
+ }
+
+ const start = async () => {
+ if (!photo || starting) return
+ setStarting(true)
+ setError('')
+ try {
+ const jobId = await llmRepo.scanReceipt(tripId, photo)
+ await saveImportFiles(jobId, [photo])
+ addTask({ id: jobId, tripId: String(tripId), label: photo.name, total: 1, files: [photo], kind: 'costs' })
+ setIsOpen(false)
+ } catch (err) {
+ const message = (err as { response?: { data?: { error?: string } } })?.response?.data?.error
+ setError(message ?? t('costs.scan.failed'))
+ } finally {
+ setStarting(false)
+ }
+ }
+
+ return { offered: canEdit && readsPhotos, isOpen, open, close, photo, choose, error, starting, start }
+}
+
+export type ReceiptScan = ReturnType
diff --git a/client/src/components/Planner/BookingImportModal.test.tsx b/client/src/components/Planner/BookingImportModal.test.tsx
index 54063390e..b32042be3 100644
--- a/client/src/components/Planner/BookingImportModal.test.tsx
+++ b/client/src/components/Planner/BookingImportModal.test.tsx
@@ -197,3 +197,38 @@ describe('BookingImportModal', () => {
expect(onClose).toHaveBeenCalledTimes(1);
});
});
+
+describe('BookingImportModal with a model that reads images', () => {
+ const jpg = () => new File(['jpeg'], 'ticket.jpg', { type: 'image/jpeg' });
+
+ it('FE-PLANNER-BOOKIMP-020: offers photos, and imports one with AI, when the model reads images', async () => {
+ let mode: string | null = null;
+ server.use(
+ http.get('/api/health/features', () => HttpResponse.json({ bookingImport: true, aiParsing: true })),
+ http.get('/api/llm/capabilities', () => HttpResponse.json({ images: true })),
+ http.post('/api/trips/4/reservations/import/booking/async', async ({ request }) => {
+ mode = (await readMultipart(request)).fields.mode ?? null;
+ return HttpResponse.json({ jobId: 'job-2' });
+ }),
+ );
+ render(
{t('reservations.import.acceptedFormats')}
+ {readsPhotos && <> {t('reservations.import.acceptedPhotos')}}
const [provider, setProvider] = useState
((cfg.provider as string) ?? 'local')
const [model, setModel] = useState((cfg.model as string) ?? '')
const [baseUrl, setBaseUrl] = useState((cfg.baseUrl as string) ?? '')
const [apiKey, setApiKey] = useState((cfg.apiKey as string) ?? '')
+ const [vision, setVision] = useState(asLlmVision(cfg.vision))
const [saving, setSaving] = useState(false)
// Local-provider model management.
@@ -458,7 +461,7 @@ function LlmParsingConfig({ addon }: { addon: Addon }) {
setSaving(true)
try {
// Send the masked sentinel unchanged so the server keeps the stored key.
- await adminApi.updateAddon(addon.id, { config: { provider, model: model.trim(), baseUrl: baseUrl.trim(), apiKey, multimodal: cfg.multimodal === true } })
+ await adminApi.updateAddon(addon.id, { config: { provider, model: model.trim(), baseUrl: baseUrl.trim(), apiKey, vision } })
toast.success('Saved')
} catch {
toast.error('Failed to save')
@@ -546,6 +549,31 @@ function LlmParsingConfig({ addon }: { addon: Addon }) {
placeholder={provider === 'anthropic' ? 'claude-opus-4-8' : provider === 'openai' ? 'gpt-4o' : 'select or pull below'}
/>
+
+
+ {LLM_VISION_MODES.map((mode) => {
+ const active = vision === mode
+ return (
+ setVision(mode)}
+ className={`flex-1 rounded-xl border px-3 py-[10px] text-[0.8125rem] font-semibold text-m-ink ${
+ active ? 'border-[color:var(--m-act)] bg-[color:var(--m-ic)]' : 'border-[color:var(--m-rowbr)]'
+ }`}
+ >
+ {t(`admin.addons.llm.vision.${mode}`)}
+
+ )
+ })}
+
+
+
+ {t(provider === 'local' ? 'admin.addons.llm.vision.hintLocal' : 'admin.addons.llm.vision.hintCloud')}
+
+
{/* Local model management (Ollama) */}
{provider === 'local' && (
diff --git a/client/src/mobile/screens/trip/MTripShell.tsx b/client/src/mobile/screens/trip/MTripShell.tsx
index cffc7685c..190761781 100644
--- a/client/src/mobile/screens/trip/MTripShell.tsx
+++ b/client/src/mobile/screens/trip/MTripShell.tsx
@@ -13,6 +13,7 @@ import MMapArea from './map/MMapArea'
import MPlacesBrowser from './places/MPlacesBrowser'
import MTripTabPanel from './tabs/MTripTabPanel'
import MTripSheets from './sheets/MTripSheets'
+import MReceiptScanButton from './tabs/MReceiptScanButton'
import MTripLoadingSplash from './MTripLoadingSplash'
import { usePluginDayTints, dayTintBackground } from '../../../components/Plugins/PluginDaySchedule'
import { stageOf } from '../../../components/Roadtrip/roadtripRowModel'
@@ -593,6 +594,7 @@ export default function MTripShell({
{trTab === 'finanzplan' && (
setAddExpenseSignal(s => s + 1)} />
+ setExportCostsCsvSignal(s => s + 1)} size={40} className="text-m-muted backdrop-blur-[24px] backdrop-saturate-[1.7]">
diff --git a/client/src/mobile/screens/trip/sheets/MCostSheet.tsx b/client/src/mobile/screens/trip/sheets/MCostSheet.tsx
index fa2e86b4e..0c3e7acd4 100644
--- a/client/src/mobile/screens/trip/sheets/MCostSheet.tsx
+++ b/client/src/mobile/screens/trip/sheets/MCostSheet.tsx
@@ -15,7 +15,7 @@ import { splitShareLabel, useExpenseFx } from '../../../../components/Budget/exp
import { SYMBOLS, SPLIT_COLORS, currenciesWith } from '../../../../components/Budget/BudgetPanel.constants'
import { COST_CATEGORY_LIST, catMeta } from '../../../../components/Budget/costsCategories'
import { localToday } from '../../../../components/Planner/today'
-import { amountPattern, calculateTicketShares, hasTicketSplit, NOTE_MAX, readTicketItems, readUserNote, splitEqualShares, writeTicketItems, type TicketItem } from '../../../../components/Budget/CostsPanel.helpers'
+import { amountPattern, calculateTicketShares, hasTicketSplit, newExpenseSeed, NOTE_MAX, readTicketItems, readUserNote, splitEqualShares, writeTicketItems, type TicketItem } from '../../../../components/Budget/CostsPanel.helpers'
import type { ExpensePrefill } from '../../../../components/Budget/CostsPanel'
import { payersBalanced, rebalancePayers } from '../../../../components/Budget/CostsPanel.helpers'
import GuestBadge from '../../../../components/shared/GuestBadge'
@@ -76,15 +76,14 @@ export default function MCostSheet({ tripId, base, people, me, editing, prefill,
const [cat, setCat] = useState(editing ? catMeta(editing.category).key : (prefill?.category || 'food'))
const [catOpen, setCatOpen] = useState(false)
const [note, setNote] = useState(() => readUserNote(editing))
- const [currency, setCurrency] = useState(editingCurrency)
- const [day, setDay] = useState(editing?.expense_date || localToday())
+ const [seed] = useState(() => newExpenseSeed(prefill, base, people.map(p => p.id), localToday()))
+ const [currency, setCurrency] = useState(editing ? editingCurrency : seed.currency)
+ const [day, setDay] = useState(editing ? (editing.expense_date || localToday()) : seed.day)
// Edit and prefill seeds are padded to the currency's decimals (#2175), same
- // as the desktop modal: a saved 4,90 must reopen as "4,90", not "4,9". A
- // prefill has no currency of its own and is read as `base`.
+ // as the desktop modal: a saved 4,90 must reopen as "4,90", not "4,9".
const [total, setTotal] = useState(() => {
if (editing) return editing.total_price ? amountToInputString(editing.total_price, editingCurrency) : ''
- if (prefill?.amount != null) return amountToInputString(prefill.amount, base)
- return ''
+ return seed.total
})
const [participants, setParticipants] = useState>(() =>
editing ? new Set((editing.members || []).map(m => m.user_id)) : new Set(people.map(p => p.id)))
@@ -115,7 +114,7 @@ export default function MCostSheet({ tripId, base, people, me, editing, prefill,
return 'equally'
})
- const [ticketItems, setTicketItems] = useState(() => readTicketItems(editing))
+ const [ticketItems, setTicketItems] = useState(() => editing ? readTicketItems(editing) : seed.ticketItems)
const [customAmounts, setCustomAmounts] = useState>(() => {
const m: Record = {}
@@ -128,7 +127,7 @@ export default function MCostSheet({ tripId, base, people, me, editing, prefill,
})
const [receipts, setReceipts] = useState(() => editing?.receipts || [])
- const [pendingReceiptFiles, setPendingReceiptFiles] = useState([])
+ const [pendingReceiptFiles, setPendingReceiptFiles] = useState(() => editing ? [] : seed.receiptFiles)
const [uploadingReceipt, setUploadingReceipt] = useState(false)
const [previewReceipts, setPreviewReceipts] = useState<{ receipts: BudgetItemReceipt[]; initialIndex: number } | null>(null)
diff --git a/client/src/mobile/screens/trip/sheets/MTripSheets.tsx b/client/src/mobile/screens/trip/sheets/MTripSheets.tsx
index c81cfa878..4f2d9c724 100644
--- a/client/src/mobile/screens/trip/sheets/MTripSheets.tsx
+++ b/client/src/mobile/screens/trip/sheets/MTripSheets.tsx
@@ -5,6 +5,7 @@ import AirTrailImportModal from '../../../../components/Planner/AirTrailImportMo
import TripFormModal from '../../../../components/Trips/TripFormModal'
import TripMembersModal from '../../../../components/Trips/TripMembersModal'
import type { ExpensePrefill } from '../../../../components/Budget/CostsPanel'
+import { expenseEditorFor } from '../../../../components/Budget/CostsPanel.helpers'
import { useAuthStore } from '../../../../store/authStore'
import { useSettingsStore } from '../../../../store/settingsStore'
import { useTripStore } from '../../../../store/tripStore'
@@ -59,6 +60,7 @@ export default function MTripSheets({ planner, shell }: MTripSheetsProps) {
else if (req.prefill) setBookingExpense({ editing: null, prefill: req.prefill })
}
const costsBase = (displayCurrency || trip?.currency || 'EUR').toUpperCase()
+ const expenseEditor = expenseEditorFor(bookingExpense, () => setBookingExpense(null), planner.receiptExpense, planner.clearReceiptExpense)
return (
<>
@@ -142,16 +144,17 @@ export default function MTripSheets({ planner, shell }: MTripSheetsProps) {
/>
)}
- {bookingExpense && (
+ {expenseEditor && (
setBookingExpense(null)}
- onSaved={() => { setBookingExpense(null); loadBudgetItems(tripId) }}
+ editing={expenseEditor.editing}
+ prefill={expenseEditor.prefill}
+ onClose={expenseEditor.close}
+ onSaved={() => { expenseEditor.close(); loadBudgetItems(tripId) }}
/>
)}
diff --git a/client/src/mobile/screens/trip/tabs/MReceiptScanButton.tsx b/client/src/mobile/screens/trip/tabs/MReceiptScanButton.tsx
new file mode 100644
index 000000000..5e2a2da91
--- /dev/null
+++ b/client/src/mobile/screens/trip/tabs/MReceiptScanButton.tsx
@@ -0,0 +1,20 @@
+import { ScanLine } from 'lucide-react'
+import MIconBtn from '../../../components/MIconBtn'
+import { useReceiptScan } from '../../../../components/Budget/useReceiptScan'
+import { ReceiptScanModal } from '../../../../components/Budget/ReceiptScanModal'
+import { useTranslation } from '../../../../i18n'
+
+/** The Costs header's "Scan receipt" button on a phone; the dialog and the scan are the shared ones. */
+export default function MReceiptScanButton({ tripId, canEdit }: { tripId: number; canEdit: boolean }) {
+ const { t } = useTranslation()
+ const scan = useReceiptScan(tripId, canEdit)
+ if (!scan.offered) return null
+ return (
+ <>
+
+
+
setBookingExpense(null)}
- onSaved={() => { setBookingExpense(null); loadBudgetItems(tripId) }}
+ editing={expenseEditor.editing}
+ prefill={expenseEditor.prefill}
+ onClose={expenseEditor.close}
+ onSaved={() => { expenseEditor.close(); loadBudgetItems(tripId) }}
/>
diff --git a/client/src/pages/tripPlanner/useTripPlanner.test.tsx b/client/src/pages/tripPlanner/useTripPlanner.test.tsx
index 6f0088be4..0d9ddce9a 100644
--- a/client/src/pages/tripPlanner/useTripPlanner.test.tsx
+++ b/client/src/pages/tripPlanner/useTripPlanner.test.tsx
@@ -17,7 +17,7 @@ import {
healthApi, airtrailApi, mapsApi,
} from '../../api/client'
import { accommodationRepo } from '../../repo/accommodationRepo'
-import { offlineDb } from '../../db/offlineDb'
+import { offlineDb, saveImportFiles, getImportFiles } from '../../db/offlineDb'
import { getCached, fetchPhoto } from '../../services/photoService'
import type { Accommodation, Place, Reservation, Settings } from '../../types'
@@ -2104,6 +2104,66 @@ describe('useTripPlanner — booking import review', () => {
})
})
+describe('useTripPlanner — a receipt scanned from Costs', () => {
+ const RECEIPT = { merchant: 'Café', date: '2026-09-20', total: 12.5, currency: 'EUR', items: [{ name: 'Tart', price: 12.5 }] }
+
+ it('FE-TP-HOOK-160: a read receipt opens the expense editor pre-filled, with the photo to attach, and clears the widget', async () => {
+ seedTrip()
+ const photo = new File(['x'], 'bill.jpg', { type: 'image/jpeg' })
+ useBackgroundTasksStore.setState({
+ tasks: [{
+ id: 'job-r', tripId: '42', label: 'bill.jpg', status: 'done', done: 1, total: 1, kind: 'costs',
+ reviewRequested: true, items: [], receipt: RECEIPT, sourceFiles: [photo],
+ }] as never,
+ })
+
+ const { result } = await renderPlanner()
+
+ await waitFor(() => expect(result.current.receiptExpense).not.toBeNull())
+ expect(result.current.receiptExpense).toEqual({
+ name: 'Café', amount: 12.5, currency: 'EUR', date: '2026-09-20', lines: [{ name: 'Tart', price: 12.5 }], receiptFiles: [photo],
+ })
+ expect(result.current.showReservationModal).toBe(false)
+ expect(useBackgroundTasksStore.getState().tasks).toHaveLength(0)
+
+ act(() => { result.current.clearReceiptExpense() })
+ expect(result.current.receiptExpense).toBeNull()
+ })
+
+ it('FE-TP-HOOK-162: after a reload the photo comes back from IndexedDB for the review', async () => {
+ seedTrip()
+ const photo = new File(['x'], 'bill.jpg', { type: 'image/jpeg' })
+ await saveImportFiles('job-db', [photo])
+ useBackgroundTasksStore.setState({
+ tasks: [{
+ id: 'job-db', tripId: '42', label: 'bill.jpg', status: 'done', done: 1, total: 1, kind: 'costs',
+ reviewRequested: true, items: [], receipt: RECEIPT,
+ }] as never,
+ })
+
+ const { result } = await renderPlanner()
+
+ await waitFor(() => expect(result.current.receiptExpense).not.toBeNull())
+ expect(result.current.receiptExpense?.receiptFiles?.map(f => f.name)).toEqual(['bill.jpg'])
+ await waitFor(async () => expect(await getImportFiles('job-db')).toEqual([]))
+ })
+
+ it('FE-TP-HOOK-161: a scan that read nothing just clears the widget', async () => {
+ seedTrip()
+ useBackgroundTasksStore.setState({
+ tasks: [{
+ id: 'job-n', tripId: '42', label: 'bill.jpg', status: 'done', done: 1, total: 1, kind: 'costs',
+ reviewRequested: true, items: [], receipt: null,
+ }] as never,
+ })
+
+ const { result } = await renderPlanner()
+
+ await waitFor(() => expect(useBackgroundTasksStore.getState().tasks).toHaveLength(0))
+ expect(result.current.receiptExpense).toBeNull()
+ })
+})
+
describe('useTripPlanner — misc state', () => {
it('FE-TP-HOOK-095: the map tile url falls back to the default basemap', async () => {
seedTrip()
diff --git a/client/src/pages/tripPlanner/useTripPlanner.ts b/client/src/pages/tripPlanner/useTripPlanner.ts
index 02613edf2..397cdf0ce 100644
--- a/client/src/pages/tripPlanner/useTripPlanner.ts
+++ b/client/src/pages/tripPlanner/useTripPlanner.ts
@@ -22,6 +22,8 @@ import { accommodationRepo } from '../../repo/accommodationRepo'
import { offlineDb, getImportFiles, deleteImportFiles } from '../../db/offlineDb'
import { isEffectivelyOffline } from '../../sync/networkMode'
import { useBackgroundTasksStore } from '../../store/backgroundTasksStore'
+import { receiptToPrefill } from '../../components/Budget/CostsPanel.helpers'
+import type { ExpensePrefill } from '../../components/Budget/CostsPanel'
import { useAuthStore } from '../../store/authStore'
import { useResizablePanels } from '../../hooks/useResizablePanels'
import { useTripWebSocket } from '../../hooks/useTripWebSocket'
@@ -435,6 +437,8 @@ export function useTripPlanner() {
const [reservationPrefill, setReservationPrefill] = useState(null)
const [transportPrefill, setTransportPrefill] = useState(null)
const [importReviewActive, setImportReviewActive] = useState(false)
+ // The expense a scanned receipt pre-fills, opened by the page's expense editor.
+ const [receiptExpense, setReceiptExpense] = useState(null)
const importQueueRef = useRef([])
// The files this import was parsed from, so each reviewed booking can attach its source doc.
const importSourceFilesRef = useRef([])
@@ -2727,13 +2731,26 @@ export function useTripPlanner() {
const task = bgTasks.find(
(tk) => tk.tripId === String(tripId) && tk.status === 'done' && tk.reviewRequested && !tk.consumed,
)
- if (task && task.items && task.items.length > 0) {
+ if (task && task.kind === 'costs') {
+ // A scanned receipt is reviewed in the expense editor, pre-filled with what
+ // was read and with the photo waiting to be attached when it is saved.
+ const receipt = task.receipt
+ const jobId = task.id
+ const inMemory = task.sourceFiles
+ dismissBgTask(jobId)
+ if (!receipt) return
+ void (async () => {
+ const files = inMemory && inMemory.length ? inMemory : await getImportFiles(jobId)
+ deleteImportFiles(jobId)
+ setReceiptExpense(receiptToPrefill(receipt, files))
+ })()
+ } else if (task && task.items && task.items.length > 0) {
// Hand the items (and the source files, to attach to each booking) to the review flow
// and clear the widget entry — once the user hit "review", the background card is done.
const items = task.items
const jobId = task.id
const inMemory = task.sourceFiles
- const kind = task.kind ?? 'bookings'
+ const kind = task.kind === 'transports' ? 'transports' : 'bookings'
dismissBgTask(jobId)
// Prefer the in-memory files (immediate path); after a reload they live in IndexedDB.
void (async () => {
@@ -2866,6 +2883,7 @@ export function useTripPlanner() {
transportModalDayId, setTransportModalDayId,
transportModalAutomated, setTransportModalAutomated, transitPrefill, setTransitPrefill, transitJourney, setTransitJourney,
reservationPrefill, transportPrefill, importReviewActive, startImportReview, advanceImportReview,
+ receiptExpense, clearReceiptExpense: () => setReceiptExpense(null),
routeShown, setRouteShown, autoShowRoute, transitRoutesShown, routeProfile, setRouteProfile, routeVias, fitKey, setFitKey,
mobileSidebarOpen, setMobileSidebarOpen, mobilePlanScrollTopRef, mobilePlacesScrollTopRef,
deletePlaceId, setDeletePlaceId, deletePlaceIds, setDeletePlaceIds, deletePlaceNote, deletePlacesNote,
diff --git a/client/src/repo/llmRepo.test.ts b/client/src/repo/llmRepo.test.ts
new file mode 100644
index 000000000..b58584a9b
--- /dev/null
+++ b/client/src/repo/llmRepo.test.ts
@@ -0,0 +1,49 @@
+// FE-REPO-LLM-001 to FE-REPO-LLM-004
+import { describe, it, expect, beforeEach } from 'vitest'
+import { http, HttpResponse } from 'msw'
+import { server } from '../../tests/helpers/msw/server'
+import { readMultipart } from '../../tests/helpers/multipart'
+import { llmRepo } from './llmRepo'
+
+function setOnline(v: boolean): void {
+ Object.defineProperty(navigator, 'onLine', { value: v, writable: true, configurable: true })
+}
+
+beforeEach(() => setOnline(true))
+
+describe('llmRepo', () => {
+ it('FE-REPO-LLM-001: readsPhotos answers what the server says the model reads', async () => {
+ server.use(http.get('/api/llm/capabilities', () => HttpResponse.json({ images: true })))
+ await expect(llmRepo.readsPhotos()).resolves.toBe(true)
+ server.use(http.get('/api/llm/capabilities', () => HttpResponse.json({ images: false })))
+ await expect(llmRepo.readsPhotos()).resolves.toBe(false)
+ })
+
+ it('FE-REPO-LLM-002: readsPhotos is false offline, without asking', async () => {
+ let asked = false
+ server.use(http.get('/api/llm/capabilities', () => { asked = true; return HttpResponse.json({ images: true }) }))
+ setOnline(false)
+ await expect(llmRepo.readsPhotos()).resolves.toBe(false)
+ expect(asked).toBe(false)
+ })
+
+ it('FE-REPO-LLM-003: scanReceipt uploads the photo as `file` and answers the job id', async () => {
+ let sent: string[] = []
+ server.use(http.post('/api/trips/7/budget/receipt-scan', async ({ request }) => {
+ sent = (await readMultipart(request)).filenames
+ return HttpResponse.json({ jobId: 'job-9' })
+ }))
+ const photo = new File(['jpeg'], 'bill.jpg', { type: 'image/jpeg' })
+ await expect(llmRepo.scanReceipt(7, photo)).resolves.toBe('job-9')
+ // One file part; jsdom names every File part 'blob' on the wire, so the name is not checked.
+ expect(sent).toHaveLength(1)
+ })
+
+ it('FE-REPO-LLM-004: scanReceipt refuses offline and rejects an answer without a job id', async () => {
+ setOnline(false)
+ await expect(llmRepo.scanReceipt(7, new File(['x'], 'a.jpg'))).rejects.toThrow(/connection/)
+ setOnline(true)
+ server.use(http.post('/api/trips/7/budget/receipt-scan', () => HttpResponse.json({})))
+ await expect(llmRepo.scanReceipt(7, new File(['x'], 'a.jpg'))).rejects.toThrow()
+ })
+})
diff --git a/client/src/repo/llmRepo.ts b/client/src/repo/llmRepo.ts
new file mode 100644
index 000000000..8bbc40de0
--- /dev/null
+++ b/client/src/repo/llmRepo.ts
@@ -0,0 +1,25 @@
+import { llmCapabilitiesResponseSchema, receiptScanStartResponseSchema } from '@trek/shared'
+import { apiClient, postMultipart } from '../api/client'
+import { isEffectivelyOffline } from '../sync/networkMode'
+
+/**
+ * What the AI model can read, and the receipt scan that depends on it. Both are
+ * online-only: the model runs on the server, so offline a photo is simply not
+ * offered.
+ */
+export const llmRepo = {
+ /** Whether a photo is worth offering to this user: their AI model reads images. */
+ async readsPhotos(): Promise {
+ if (isEffectivelyOffline()) return false
+ const reply = await apiClient.get('/llm/capabilities')
+ return llmCapabilitiesResponseSchema.parse(reply.data).images
+ },
+
+ /** Start reading one receipt photo in the background; answers the import job id. */
+ async scanReceipt(tripId: number | string, photo: File): Promise {
+ if (isEffectivelyOffline()) throw new Error('Scanning a receipt requires a connection')
+ const form = new FormData()
+ form.append('file', photo)
+ return receiptScanStartResponseSchema.parse(await postMultipart(`/trips/${tripId}/budget/receipt-scan`, form)).jobId
+ },
+}
diff --git a/client/src/store/backgroundTasksStore.test.ts b/client/src/store/backgroundTasksStore.test.ts
new file mode 100644
index 000000000..c371f53af
--- /dev/null
+++ b/client/src/store/backgroundTasksStore.test.ts
@@ -0,0 +1,29 @@
+import { describe, it, expect, beforeEach } from 'vitest'
+import { useBackgroundTasksStore, taskFoundSomething, type BackgroundImportTask } from './backgroundTasksStore'
+
+const RECEIPT = { merchant: 'Café', date: null, total: 4, currency: 'EUR', items: [] }
+
+beforeEach(() => useBackgroundTasksStore.setState({ tasks: [] }))
+
+describe('backgroundTasksStore with a receipt scan', () => {
+ it('keeps what the scan read on the finished task', () => {
+ const s = useBackgroundTasksStore.getState()
+ s.addTask({ id: 'j1', tripId: '7', label: 'bill.jpg', total: 1, kind: 'costs' })
+ s.setDone('j1', '7', [], [], RECEIPT)
+ expect(useBackgroundTasksStore.getState().tasks[0]).toMatchObject({ status: 'done', kind: 'costs', receipt: RECEIPT, items: [] })
+ })
+})
+
+describe('taskFoundSomething', () => {
+ const task = (over: Partial): BackgroundImportTask => ({ id: 'j', tripId: '7', label: 'x', status: 'done', done: 1, total: 1, ...over })
+
+ it('asks a receipt scan for its receipt', () => {
+ expect(taskFoundSomething(task({ kind: 'costs', items: [], receipt: RECEIPT }))).toBe(true)
+ expect(taskFoundSomething(task({ kind: 'costs', items: [], receipt: null }))).toBe(false)
+ })
+
+ it('asks a booking parse for its items', () => {
+ expect(taskFoundSomething(task({ kind: 'bookings', items: [{} as never] }))).toBe(true)
+ expect(taskFoundSomething(task({ items: [] }))).toBe(false)
+ })
+})
diff --git a/client/src/store/backgroundTasksStore.ts b/client/src/store/backgroundTasksStore.ts
index bf280f715..f0861e09b 100644
--- a/client/src/store/backgroundTasksStore.ts
+++ b/client/src/store/backgroundTasksStore.ts
@@ -1,6 +1,6 @@
import { create } from 'zustand'
import { persist } from 'zustand/middleware'
-import type { BookingImportPreviewItem, BookingImportMode } from '@trek/shared'
+import type { BookingImportPreviewItem, BookingImportMode, ReceiptRead } from '@trek/shared'
/**
* Tracks booking-import parses that run in the BACKGROUND (the async endpoint).
@@ -15,6 +15,8 @@ import type { BookingImportPreviewItem, BookingImportMode } from '@trek/shared'
* on mount. We deliberately persist neither the parsed `items` (re-fetched) nor the
* transient review flags (so a reload never auto-reopens the review flow).
*/
+export type BackgroundTaskKind = 'transports' | 'bookings' | 'costs'
+
export interface BackgroundImportTask {
id: string // server job id
tripId: string
@@ -23,6 +25,8 @@ export interface BackgroundImportTask {
done: number
total: number
items?: BookingImportPreviewItem[]
+ /** A receipt scanned from Costs (`kind: 'costs'`): what the model read, or null when it read nothing. */
+ receipt?: ReceiptRead | null
warnings?: string[]
error?: string
reviewRequested?: boolean // user clicked "review" — the trip page consumes it
@@ -37,16 +41,17 @@ export interface BackgroundImportTask {
* when its type is guessed (#2076), and unlike the other transient flags it IS
* persisted: the parse deliberately outlives navigation and reload, and the
* review is triggered by the global widget on any page, so a component-local
- * value is gone by the time it is needed.
+ * value is gone by the time it is needed. `costs` is a receipt scan, reviewed
+ * in the expense editor instead of the booking forms.
*/
- kind?: 'transports' | 'bookings'
+ kind?: BackgroundTaskKind
}
interface BackgroundTasksState {
tasks: BackgroundImportTask[]
- addTask: (task: { id: string; tripId: string; label: string; total: number; files?: File[]; mode?: BookingImportMode; kind?: 'transports' | 'bookings' }) => void
+ addTask: (task: { id: string; tripId: string; label: string; total: number; files?: File[]; mode?: BookingImportMode; kind?: BackgroundTaskKind }) => void
setProgress: (id: string, tripId: string, done: number, total: number) => void
- setDone: (id: string, tripId: string, items: BookingImportPreviewItem[], warnings: string[]) => void
+ setDone: (id: string, tripId: string, items: BookingImportPreviewItem[], warnings: string[], receipt?: ReceiptRead | null) => void
setError: (id: string, tripId: string, error: string) => void
requestReview: (id: string) => void
markConsumed: (id: string) => void
@@ -73,7 +78,7 @@ export const useBackgroundTasksStore = create()(
tasks: [],
addTask: ({ id, tripId, label, total, files, mode, kind }) => upsert(id, tripId, { label, total, status: 'running', done: 0, sourceFiles: files, mode, kind }),
setProgress: (id, tripId, done, total) => upsert(id, tripId, { done, total, status: 'running' }),
- setDone: (id, tripId, items, warnings) => upsert(id, tripId, { status: 'done', items, warnings, done: items?.length ?? 0 }),
+ setDone: (id, tripId, items, warnings, receipt) => upsert(id, tripId, { status: 'done', items, warnings, receipt, done: items?.length ?? 0 }),
setError: (id, tripId, error) => upsert(id, tripId, { status: 'error', error }),
requestReview: (id) => set((s) => ({ tasks: s.tasks.map((t) => (t.id === id ? { ...t, reviewRequested: true } : t)) })),
markConsumed: (id) => set((s) => ({ tasks: s.tasks.map((t) => (t.id === id ? { ...t, consumed: true, reviewRequested: false } : t)) })),
@@ -96,3 +101,8 @@ export const useBackgroundTasksStore = create()(
},
),
)
+
+/** Whether a finished task has something to review: a read receipt, or at least one booking. */
+export function taskFoundSomething(task: BackgroundImportTask): boolean {
+ return task.kind === 'costs' ? Boolean(task.receipt) : (task.items?.length ?? 0) > 0
+}
diff --git a/client/tests/helpers/msw/handlers/addons.ts b/client/tests/helpers/msw/handlers/addons.ts
index 8e2f84687..aa970bbcc 100644
--- a/client/tests/helpers/msw/handlers/addons.ts
+++ b/client/tests/helpers/msw/handlers/addons.ts
@@ -1,6 +1,8 @@
import { http, HttpResponse } from 'msw';
export const addonHandlers = [
+ // Whose AI model reads images: off unless a test says otherwise, so no photo picker shows by default.
+ http.get('/api/llm/capabilities', () => HttpResponse.json({ images: false })),
http.get('/api/addons', () => {
return HttpResponse.json({
bagTracking: false,
diff --git a/client/tests/unit/mobile/admin/MAdminAddonManager.test.tsx b/client/tests/unit/mobile/admin/MAdminAddonManager.test.tsx
index 970645c3e..466fe9875 100644
--- a/client/tests/unit/mobile/admin/MAdminAddonManager.test.tsx
+++ b/client/tests/unit/mobile/admin/MAdminAddonManager.test.tsx
@@ -649,7 +649,7 @@ describe('MAdminAddonManager', () => {
model: 'gpt-4o',
baseUrl: 'https://api.openai.com/v1',
apiKey: '••••••••',
- multimodal: true,
+ vision: 'auto',
},
});
@@ -657,6 +657,30 @@ describe('MAdminAddonManager', () => {
await screen.findByText('Failed to save');
});
+ it('FE-MOB-AADD-030: whether the model reads images is a three-way choice, saved as picked', async () => {
+ const user = userEvent.setup();
+ const bodies: unknown[] = [];
+ server.use(
+ addonsRoute([llmAddon({ provider: 'local', model: 'llava:7b', baseUrl: '', apiKey: '', vision: 'on' })]),
+ http.get('/api/admin/llm/local/models', () => HttpResponse.json({ models: [] })),
+ http.put('/api/admin/addons/llm_parsing', async ({ request }) => {
+ bodies.push(await request.json());
+ return HttpResponse.json({ success: true });
+ }),
+ );
+ render(<> {
+ return this.llmParse.readsImages(userId);
+ }
+
/**
* Parse uploaded files and return a preview list. Does NOT persist anything.
* Runs kitinerary first; depending on `mode`, falls back to the LLM:
@@ -155,8 +161,10 @@ export class BookingImportService {
let kiItems: KiReservation[] = [];
let aiUsed = false;
- // Stage 1: kitinerary (skipped entirely when forcing AI).
- if (mode !== 'force-ai' && kitineraryAvailable) {
+ // Stage 1: kitinerary (skipped entirely when forcing AI, and for a photo,
+ // which only a model can read).
+ const photo = imageMimeType(file.originalname) !== null;
+ if (mode !== 'force-ai' && kitineraryAvailable && !photo) {
try {
kiItems = await this.extractor.extract(file.buffer, file.originalname);
} catch (err) {
diff --git a/server/src/nest/booking-import/import-jobs.service.ts b/server/src/nest/booking-import/import-jobs.service.ts
index 5aaeb65b1..cb709b81f 100644
--- a/server/src/nest/booking-import/import-jobs.service.ts
+++ b/server/src/nest/booking-import/import-jobs.service.ts
@@ -2,7 +2,8 @@ import { Injectable } from '@nestjs/common';
import { randomUUID } from 'node:crypto';
import { RealtimeService } from '../realtime/realtime.service';
import { BookingImportService } from './booking-import.service';
-import type { BookingImportMode, BookingImportPreviewResponse, TrekWsPayload } from '@trek/shared';
+import { LlmParseService } from '../llm-parse/llm-parse.service';
+import type { BookingImportMode, BookingImportPreviewResponse, ReceiptScanResult, TrekWsPayload } from '@trek/shared';
type JobStatus = 'running' | 'done' | 'error';
@@ -13,7 +14,7 @@ interface ImportJob {
status: JobStatus;
done: number;
total: number;
- result?: BookingImportPreviewResponse;
+ result?: BookingImportPreviewResponse | ReceiptScanResult;
error?: string;
createdAt: number;
}
@@ -28,7 +29,9 @@ const JOB_TTL_MS = 10 * 60_000;
* user's sockets via `broadcastToUser` (which reaches them on ANY page, not just the
* trip room). This is what lets the upload modal close at once and a background widget
* track the work while the user keeps navigating. The actual parsing is the same
- * `BookingImportService.preview` the synchronous endpoint uses.
+ * `BookingImportService.preview` the synchronous endpoint uses. A receipt scanned
+ * from the Costs tab rides the same queue, with `LlmParseService.readReceipt` as
+ * its work: it is the same slow model on the same server.
*/
@Injectable()
export class ImportJobsService {
@@ -36,17 +39,44 @@ export class ImportJobsService {
/** Tail of each user's job chain — parses run one at a time per user, not all at once. */
private readonly chains = new Map>();
- constructor(private readonly bookingImport: BookingImportService, private readonly realtime: RealtimeService) {}
+ constructor(
+ private readonly bookingImport: BookingImportService,
+ private readonly realtime: RealtimeService,
+ private readonly llmParse: LlmParseService,
+ ) {}
/** Create a job and queue it behind the user's other parses; returns the job id at once. */
start(tripId: string, files: Express.Multer.File[], mode: BookingImportMode, userId: number): string {
+ return this.enqueue(tripId, userId, files.length, (job) =>
+ this.bookingImport.preview(files, mode, job.userId, (done, total, fileName) => {
+ job.done = done;
+ this.push(job, 'import:progress', { status: 'running', done, total, fileName });
+ }),
+ );
+ }
+
+ /** Queue the read of one receipt photo, behind the same user's other parses. */
+ startReceipt(tripId: string, file: Express.Multer.File, userId: number): string {
+ return this.enqueue(tripId, userId, 1, async (job) => {
+ const result = await this.llmParse.readReceipt({ buffer: file.buffer, originalName: file.originalname }, job.userId);
+ job.done = 1;
+ return result;
+ });
+ }
+
+ private enqueue(
+ tripId: string,
+ userId: number,
+ total: number,
+ work: (job: ImportJob) => Promise,
+ ): string {
const id = randomUUID();
- const job: ImportJob = { id, tripId, userId, status: 'running', done: 0, total: files.length, createdAt: Date.now() };
+ const job: ImportJob = { id, tripId, userId, status: 'running', done: 0, total, createdAt: Date.now() };
this.jobs.set(id, job);
// Chain onto the user's previous parse so they run sequentially (one CPU-heavy
// inference at a time), while the request returns immediately.
const prev = this.chains.get(userId) ?? Promise.resolve();
- const next = prev.then(() => this.run(job, files, mode)).catch(() => {});
+ const next = prev.then(() => this.run(job, work)).catch(() => {});
this.chains.set(userId, next);
void next.finally(() => {
if (this.chains.get(userId) === next) this.chains.delete(userId);
@@ -59,13 +89,10 @@ export class ImportJobsService {
return job && job.userId === userId ? job : undefined;
}
- private async run(job: ImportJob, files: Express.Multer.File[], mode: BookingImportMode): Promise {
+ private async run(job: ImportJob, work: (job: ImportJob) => Promise): Promise {
this.push(job, 'import:progress', { status: 'running', done: 0, total: job.total });
try {
- const result = await this.bookingImport.preview(files, mode, job.userId, (done, total, fileName) => {
- job.done = done;
- this.push(job, 'import:progress', { status: 'running', done, total, fileName });
- });
+ const result = await work(job);
job.status = 'done';
job.result = result;
this.push(job, 'import:done', { result });
diff --git a/server/src/nest/llm-parse/clients/anthropic.client.ts b/server/src/nest/llm-parse/clients/anthropic.client.ts
index be0600f2f..63c120c33 100644
--- a/server/src/nest/llm-parse/clients/anthropic.client.ts
+++ b/server/src/nest/llm-parse/clients/anthropic.client.ts
@@ -6,13 +6,14 @@ import { UnreadableLlmResponse } from './openai-compatible.client';
const MAX_TOKENS = 8192;
const ANTHROPIC_VERSION = '2023-06-01';
-const TOOL_NAME = 'emit_reservations';
/**
* Anthropic Messages API client. Structured output via forced tool-use: a single
- * `emit_reservations` tool whose `input_schema` is the reservations schema, with
- * `tool_choice` forcing it — the documented, reliable way to get structured JSON.
- * PDFs go as native base64 `document` blocks (Anthropic reads scanned PDFs).
+ * `emit_` tool (`emit_reservations` by default) whose `input_schema` is
+ * the caller's schema, with `tool_choice` forcing it — the documented, reliable
+ * way to get structured JSON.
+ * PDFs go as native base64 `document` blocks (Anthropic reads scanned PDFs), photos
+ * as `image` blocks.
* Raw fetch (no SDK) to match the codebase's HTTP style.
*/
export class AnthropicClient implements LlmExtractionClient {
@@ -24,16 +25,21 @@ export class AnthropicClient implements LlmExtractionClient {
const base = (input.baseUrl ?? 'https://api.anthropic.com').replace(/(? }[];
};
if (data.stop_reason === 'refusal') {
@@ -95,9 +101,9 @@ export class AnthropicClient implements LlmExtractionClient {
);
}
- const toolUse = data.content?.find(b => b.type === 'tool_use' && b.name === TOOL_NAME);
- if (!toolUse) throw new UnreadableLlmResponse(`the model answered without calling ${TOOL_NAME}`);
- return toReservationList(toolUse.input?.reservations);
+ const toolUse = data.content?.find(b => b.type === 'tool_use' && b.name === toolName);
+ if (!toolUse) throw new UnreadableLlmResponse(`the model answered without calling ${toolName}`);
+ return toReservationList(toolUse.input?.[rootKey], rootKey);
}
}
diff --git a/server/src/nest/llm-parse/clients/nuextract.ts b/server/src/nest/llm-parse/clients/nuextract.ts
index 372f792e7..ce4114edc 100644
--- a/server/src/nest/llm-parse/clients/nuextract.ts
+++ b/server/src/nest/llm-parse/clients/nuextract.ts
@@ -110,7 +110,7 @@ function clean(value: unknown): unknown {
* treating the right-most separator as the decimal point. Returns null when there
* is no parseable amount.
*/
-function parseAmount(raw: unknown): number | null {
+export function parseAmount(raw: unknown): number | null {
if (typeof raw === 'number') return Number.isFinite(raw) ? raw : null;
if (typeof raw !== 'string') return null;
let s = raw.replace(/[^\d.,]/g, '');
diff --git a/server/src/nest/llm-parse/clients/openai-compatible.client.ts b/server/src/nest/llm-parse/clients/openai-compatible.client.ts
index f00ebec27..45ac43d75 100644
--- a/server/src/nest/llm-parse/clients/openai-compatible.client.ts
+++ b/server/src/nest/llm-parse/clients/openai-compatible.client.ts
@@ -84,18 +84,21 @@ export class OpenAiCompatibleClient implements LlmExtractionClient {
const base = (input.baseUrl ?? 'https://api.openai.com/v1').replace(/(? {
@@ -216,9 +219,9 @@ function parseNuExtract(content: string | undefined | null): Record[] {
- return toReservationList(readAnswer(content));
+/** Tolerant parse: strip code fences, JSON(5).parse, pull the `rootKey` array. */
+function parseReservations(content: string | undefined | null, rootKey: string): Record[] {
+ return toReservationList(readAnswer(content), rootKey);
}
/**
diff --git a/server/src/nest/llm-parse/image-input.ts b/server/src/nest/llm-parse/image-input.ts
new file mode 100644
index 000000000..75f7dc148
--- /dev/null
+++ b/server/src/nest/llm-parse/image-input.ts
@@ -0,0 +1,67 @@
+import { extname } from 'node:path';
+import { Jimp } from 'jimp';
+import { PDFParse } from 'pdf-parse';
+import { LLM_PHOTO_EXTENSIONS } from '@trek/shared';
+
+/** A photographed document on its way to a vision model, by the extensions shared/ lists. */
+const IMAGE_MIME_BY_EXT: Record<(typeof LLM_PHOTO_EXTENSIONS)[number], string> = {
+ '.jpg': 'image/jpeg',
+ '.jpeg': 'image/jpeg',
+ '.png': 'image/png',
+ '.webp': 'image/webp',
+};
+
+export const IMAGE_EXTENSIONS: readonly string[] = LLM_PHOTO_EXTENSIONS;
+
+/** The image MIME type for a file name, or null when it is not a photo TREK sends. */
+export function imageMimeType(fileName: string): string | null {
+ return (IMAGE_MIME_BY_EXT as Record)[extname(fileName).toLowerCase()] ?? null;
+}
+
+/**
+ * Long edge in pixels. A phone photo is around twelve megapixels, which a model
+ * pays for in prompt tokens without reading the print any better: the same
+ * receipt on a local qwen3.5:4b took 252s whole and 37-124s at this size.
+ */
+export const IMAGE_MAX_EDGE = 1600;
+const IMAGE_JPEG_QUALITY = 82;
+
+/**
+ * The image, no larger than the model needs. Returns the bytes unchanged when
+ * they are already small enough, or when they cannot be decoded (Jimp has no
+ * WebP): a photo that might be read beats one refused over a failed resize.
+ */
+export async function capImage(data: Buffer, mimeType: string): Promise<{ data: Buffer; mimeType: string }> {
+ try {
+ const image = await Jimp.read(data);
+ if (Math.max(image.width, image.height) <= IMAGE_MAX_EDGE) return { data, mimeType };
+ if (image.width >= image.height) image.resize({ w: IMAGE_MAX_EDGE });
+ else image.resize({ h: IMAGE_MAX_EDGE });
+ const resized = await image.getBuffer('image/jpeg', { quality: IMAGE_JPEG_QUALITY });
+ return { data: Buffer.from(resized), mimeType: 'image/jpeg' };
+ } catch {
+ return { data, mimeType };
+ }
+}
+
+/**
+ * Pages of a PDF read the only way a scan can be: as pictures. The booking
+ * details sit on the first pages, and every page costs the model a photo's
+ * worth of prompt, so only the first two are drawn.
+ */
+export const PDF_PAGES_AS_IMAGES = 2;
+
+/** The first pages of a PDF drawn as images, capped like a photo; empty when none could be drawn. */
+export async function renderPdfPages(data: Buffer): Promise<{ data: Buffer; mimeType: string }[]> {
+ const parser = new PDFParse({ data: new Uint8Array(data) });
+ try {
+ const shot = await parser.getScreenshot({ first: PDF_PAGES_AS_IMAGES, desiredWidth: IMAGE_MAX_EDGE, imageBuffer: true, imageDataUrl: false });
+ const pages: { data: Buffer; mimeType: string }[] = [];
+ for (const page of shot.pages) {
+ if (page.data?.length) pages.push(await capImage(Buffer.from(page.data), 'image/png'));
+ }
+ return pages;
+ } finally {
+ await parser.destroy().catch(() => {});
+ }
+}
diff --git a/server/src/nest/llm-parse/lenient-json.ts b/server/src/nest/llm-parse/lenient-json.ts
index 109683564..0203fc8f6 100644
--- a/server/src/nest/llm-parse/lenient-json.ts
+++ b/server/src/nest/llm-parse/lenient-json.ts
@@ -60,13 +60,16 @@ export function parseLenientJson(content: string | undefined | null): unknown {
* would have been accepted inside an array is accepted alone rather than thrown
* away without a word (#2375). It is still not guessing: an object without that
* key stays refused.
+ *
+ * `rootKey` names the wrapper for a caller that asks for another list than
+ * reservations (the receipt read asks for `receipts`).
*/
-export function toReservationList(value: unknown): Record[] {
+export function toReservationList(value: unknown, rootKey = 'reservations'): Record[] {
const list = (v: unknown): Record[] | null => {
if (Array.isArray(v)) return v as Record[];
if (v && typeof v === 'object') {
- const o = v as { reservations?: unknown; '@type'?: unknown };
- if (Array.isArray(o.reservations)) return o.reservations as Record[];
+ const o = v as Record;
+ if (Array.isArray(o[rootKey])) return o[rootKey] as Record[];
if (typeof o['@type'] === 'string') return [v as Record];
}
return null;
diff --git a/server/src/nest/llm-parse/llm-capabilities.controller.ts b/server/src/nest/llm-parse/llm-capabilities.controller.ts
new file mode 100644
index 000000000..6ff0b36e9
--- /dev/null
+++ b/server/src/nest/llm-parse/llm-capabilities.controller.ts
@@ -0,0 +1,22 @@
+import { Controller, Get, UseGuards } from '@nestjs/common';
+import type { LlmCapabilitiesResponse } from '@trek/shared';
+import type { User } from '../../types';
+import { JwtAuthGuard } from '../auth/jwt-auth.guard';
+import { CurrentUser } from '../auth/current-user.decorator';
+import { LlmParseService } from './llm-parse.service';
+
+/**
+ * What the caller's AI model can be handed. Per user, unlike the public
+ * /api/health/features: whose model applies depends on who is asking. The
+ * photo pickers read it to decide whether a photo is offered at all.
+ */
+@Controller('api/llm')
+@UseGuards(JwtAuthGuard)
+export class LlmCapabilitiesController {
+ constructor(private readonly llmParse: LlmParseService) {}
+
+ @Get('capabilities')
+ async capabilities(@CurrentUser() user: User): Promise {
+ return { images: await this.llmParse.readsImages(user.id) };
+ }
+}
diff --git a/server/src/nest/llm-parse/llm-config.resolver.ts b/server/src/nest/llm-parse/llm-config.resolver.ts
index 0e428f9b5..7f8f7b072 100644
--- a/server/src/nest/llm-parse/llm-config.resolver.ts
+++ b/server/src/nest/llm-parse/llm-config.resolver.ts
@@ -1,5 +1,6 @@
import { ADDON_IDS } from '../../addons';
import { AddonsService } from '../addons/addons.service';
+import { asLlmVision } from '@trek/shared';
import { decryptLlmApiKey, LLM_PROVIDERS, type LlmProvider, type ResolvedLlmConfig } from './llm-config';
import { DatabaseService } from '../database/database.service';
import { SettingsService } from '../settings/settings.service';
@@ -53,7 +54,7 @@ export class LlmConfigResolver {
model,
baseUrl: typeof cfg.baseUrl === 'string' && cfg.baseUrl.trim() ? cfg.baseUrl.trim() : undefined,
apiKey: decryptLlmApiKey(cfg.apiKey),
- multimodal: cfg.multimodal === true,
+ vision: asLlmVision(cfg.vision),
};
}
@@ -88,7 +89,9 @@ export class LlmConfigResolver {
model,
baseUrl,
apiKey,
- multimodal: settings.llm_multimodal === true,
+ // The personal switch has only ever been on or off, and was saved as the
+ // person set it, so it is read as exactly that.
+ vision: settings.llm_multimodal === true ? 'on' : 'off',
};
}
}
diff --git a/server/src/nest/llm-parse/llm-config.ts b/server/src/nest/llm-parse/llm-config.ts
index 54bdbe08a..ea28de9a0 100644
--- a/server/src/nest/llm-parse/llm-config.ts
+++ b/server/src/nest/llm-parse/llm-config.ts
@@ -1,3 +1,4 @@
+import { asLlmVision, type LlmVision } from '@trek/shared';
import { maybe_encrypt_api_key, decrypt_api_key } from '../common/crypto/apiKeyCrypto';
/**
@@ -21,7 +22,7 @@ export interface ResolvedLlmConfig {
model: string;
baseUrl?: string;
apiKey?: string;
- multimodal: boolean;
+ vision: LlmVision;
}
/** Shape of the admin instance config stored in `addons.config` (apiKey encrypted). */
@@ -30,7 +31,11 @@ export interface LlmAddonConfig {
model?: string;
baseUrl?: string;
apiKey?: string;
- multimodal?: boolean;
+ /**
+ * Replaces `multimodal`, which the admin screens wrote as `false` on every
+ * save whatever the model: a stored `false` means nothing, so it is not read.
+ */
+ vision?: LlmVision;
}
export const LLM_PROVIDERS: LlmProvider[] = ['local', 'openai', 'anthropic'];
@@ -46,6 +51,7 @@ export function prepareLlmAddonConfigForWrite(
existingStored: Record | undefined,
): Record {
const out: Record = { ...incoming };
+ if ('vision' in incoming) out.vision = asLlmVision(incoming.vision);
const key = incoming.apiKey;
if (key === undefined || key === null || key === '' || key === MASKED_VALUE) {
// Keep the existing encrypted key untouched (mask echoed or no key supplied).
diff --git a/server/src/nest/llm-parse/llm-local.service.ts b/server/src/nest/llm-parse/llm-local.service.ts
index b70ee6afb..77e6f3a5a 100644
--- a/server/src/nest/llm-parse/llm-local.service.ts
+++ b/server/src/nest/llm-parse/llm-local.service.ts
@@ -1,5 +1,17 @@
import { Injectable, HttpException } from '@nestjs/common';
+import { z } from 'zod';
import { safeFetchLlm } from '../../utils/ssrfGuard';
+import { readCappedJson } from '../../utils/cappedFetch';
+
+/** `/api/show` carries the licence and the template: tens of KB, never megabytes. */
+const MAX_SHOW_BYTES = 1024 * 1024;
+const SHOW_TIMEOUT_MS = 5_000;
+/** A model's capabilities only change when it is re-pulled; ten minutes of staleness is harmless. */
+const CAPABILITIES_TTL_MS = 10 * 60_000;
+/** An unreachable server is asked again soon, so a restart does not cost ten minutes. */
+const CAPABILITIES_MISS_TTL_MS = 60_000;
+
+const showResponseSchema = z.object({ capabilities: z.array(z.string()).optional() });
/**
* Admin helpers for managing a local OpenAI-compatible LLM server (Ollama).
@@ -11,6 +23,8 @@ import { safeFetchLlm } from '../../utils/ssrfGuard';
*/
@Injectable()
export class LlmLocalService {
+ private readonly capabilityCache = new Map();
+
/** Derive the Ollama root from a configured base URL (strip a trailing /v1). */
ollamaRoot(baseUrl: string | undefined): string {
const raw = (baseUrl ?? 'http://localhost:11434').trim();
@@ -65,4 +79,42 @@ export class LlmLocalService {
if (!res.ok || !res.body) throw new HttpException({ error: `Pull failed (${res.status})` }, 502);
return res.body;
}
+
+ /**
+ * What the local server says the model can do (`['completion', 'vision', …]`),
+ * or null when it cannot say: a bad URL, a server that is down, a model it
+ * does not have, an answer that is not what Ollama sends. Never throws, since
+ * every caller treats "unknown" as "no", and remembers the answer per server
+ * and model so a page load does not become a request to the model server.
+ */
+ async modelCapabilities(baseUrl: string | undefined, model: string): Promise {
+ const now = Date.now();
+ const key = `${baseUrl ?? ''}\n${model}`;
+ const cached = this.capabilityCache.get(key);
+ if (cached && cached.expires > now) return cached.capabilities;
+
+ let capabilities: string[] | null = null;
+ try {
+ const root = this.ollamaRoot(baseUrl);
+ const res = await safeFetchLlm(`${root}/api/show`, {
+ method: 'POST',
+ headers: { 'content-type': 'application/json' },
+ body: JSON.stringify({ model }),
+ signal: AbortSignal.timeout(SHOW_TIMEOUT_MS),
+ });
+ if (res.ok) {
+ const parsed = showResponseSchema.safeParse(await readCappedJson(res, MAX_SHOW_BYTES));
+ if (parsed.success) capabilities = parsed.data.capabilities ?? [];
+ } else {
+ void res.body?.cancel().catch(() => {});
+ }
+ } catch {
+ capabilities = null;
+ }
+ this.capabilityCache.set(key, {
+ capabilities,
+ expires: now + (capabilities ? CAPABILITIES_TTL_MS : CAPABILITIES_MISS_TTL_MS),
+ });
+ return capabilities;
+ }
}
diff --git a/server/src/nest/llm-parse/llm-parse.module.ts b/server/src/nest/llm-parse/llm-parse.module.ts
index 0b6ecfb8f..676370c18 100644
--- a/server/src/nest/llm-parse/llm-parse.module.ts
+++ b/server/src/nest/llm-parse/llm-parse.module.ts
@@ -2,6 +2,7 @@ import { Module } from '@nestjs/common';
import { LlmParseService } from './llm-parse.service';
import { LlmLocalService } from './llm-local.service';
import { LlmLocalController } from './llm-local.controller';
+import { LlmCapabilitiesController } from './llm-capabilities.controller';
import { LlmConfigResolver } from './llm-config.resolver';
import { SettingsModule } from '../settings/settings.module';
import { AddonsModule } from '../addons/addons.module';
@@ -13,7 +14,7 @@ import { AppConfigModule } from '../app-config/app-config.module';
*/
@Module({
imports: [AppConfigModule, SettingsModule, AddonsModule],
- controllers: [LlmLocalController],
+ controllers: [LlmLocalController, LlmCapabilitiesController],
providers: [LlmParseService, LlmLocalService, LlmConfigResolver],
exports: [LlmParseService, LlmConfigResolver],
})
diff --git a/server/src/nest/llm-parse/llm-parse.service.ts b/server/src/nest/llm-parse/llm-parse.service.ts
index b83c817ae..d06ebf592 100644
--- a/server/src/nest/llm-parse/llm-parse.service.ts
+++ b/server/src/nest/llm-parse/llm-parse.service.ts
@@ -1,13 +1,25 @@
import type { KiReservation } from '../booking-import/kitinerary.types';
import { createLlmClient } from './llm-client.factory';
import { LlmConfigResolver } from './llm-config.resolver';
+import type { ResolvedLlmConfig } from './llm-config';
+import { LlmLocalService } from './llm-local.service';
+import { capImage, imageMimeType, renderPdfPages } from './image-input';
+import { extractEnforced } from './router/ollama-format.client';
+import {
+ buildReceiptPrompt,
+ RECEIPT_JSON_SCHEMA,
+ RECEIPT_LIST_JSON_SCHEMA,
+ RECEIPT_ROOT_KEY,
+ RECEIPT_USER_TEXT,
+ toReceiptRead,
+} from './receipt-read';
import { buildSystemPrompt, KI_RESERVATION_JSON_SCHEMA } from './llm-prompt';
import type { LlmExtractionInput } from './llm-provider.interface';
import { isPdf, extractText } from './text-extract';
-import { routeExtraction, detectFlightNumbers, extractTotalPrice } from './router/extraction-router';
+import { routeExtraction, routeImageExtraction, detectFlightNumbers, extractTotalPrice } from './router/extraction-router';
import { toIsoCurrency } from './currency-code';
import { Injectable } from '@nestjs/common';
-import { kiReservationSchema } from '@trek/shared';
+import { kiReservationSchema, type ReceiptScanResult } from '@trek/shared';
import { RuntimeEnvService } from '../app-config/runtime-env.service';
const MIME_BY_EXT: Record = {
@@ -21,7 +33,8 @@ export interface LlmParseResult {
/**
* Orchestrates the LLM fallback: resolve config → pick client → build input
- * (native bytes vs extracted text by the `multimodal` flag) → call provider →
+ * (a photo as an image when the model reads one, a PDF natively for Anthropic,
+ * extracted text otherwise) → call provider →
* validate the response → return schema.org `KiReservation[]` for the shared
* mapper. Never throws for content/provider reasons — degrades to `[]` + a
* warning, mirroring the kitinerary extractor's tolerance.
@@ -31,6 +44,7 @@ export class LlmParseService {
constructor(
private readonly llmConfig: LlmConfigResolver,
private readonly env: RuntimeEnvService,
+ private readonly local: LlmLocalService,
) {}
/** True when the addon is enabled AND a usable config resolves for this user. */
@@ -38,6 +52,77 @@ export class LlmParseService {
return this.llmConfig.resolve(userId) !== null;
}
+ /** Whether a photo this user uploads can be read: AI parsing is set up and its model takes images. */
+ async readsImages(userId: number): Promise {
+ const config = this.llmConfig.resolve(userId);
+ return config ? this.visionFor(config) : false;
+ }
+
+ /**
+ * Read a photographed receipt for the Costs tab. Like parse(), it never throws
+ * for content or provider reasons: a refusal comes back as a null receipt and
+ * a warning naming the file.
+ */
+ async readReceipt(file: { buffer: Buffer; originalName: string }, userId: number): Promise {
+ const config = this.llmConfig.resolve(userId);
+ if (!config) return { receipt: null, warnings: ['AI parsing is not configured'] };
+ const imageType = imageMimeType(file.originalName);
+ if (!imageType) return { receipt: null, warnings: [`${file.originalName}: not a photo`] };
+ if (!(await this.visionFor(config))) {
+ return { receipt: null, warnings: [`${file.originalName}: the configured model does not read images`] };
+ }
+
+ const image = await capImage(file.buffer, imageType);
+ const prompt = buildReceiptPrompt();
+ let raw: unknown;
+ try {
+ if (config.provider === 'local') {
+ // Ollama's own chat API, as the booking router uses: `think: false` and
+ // `num_ctx` exist there and not on /v1, and a hybrid model that is left
+ // to reason spends the whole token budget on it and answers nothing.
+ raw = await extractEnforced({
+ baseUrl: config.baseUrl ?? 'http://localhost:11434/v1',
+ model: config.model,
+ apiKey: config.apiKey,
+ system: prompt,
+ user: RECEIPT_USER_TEXT,
+ schema: RECEIPT_JSON_SCHEMA,
+ images: [image.data.toString('base64')],
+ numCtx: 16384,
+ // A supermarket receipt is a long list of lines.
+ numPredict: 2048,
+ });
+ } else {
+ const list = await createLlmClient(config).extract({
+ prompt,
+ jsonSchema: RECEIPT_LIST_JSON_SCHEMA,
+ rootKey: RECEIPT_ROOT_KEY,
+ userText: RECEIPT_USER_TEXT,
+ model: config.model,
+ baseUrl: config.baseUrl,
+ apiKey: config.apiKey,
+ file: image,
+ });
+ raw = list[0];
+ }
+ } catch (err) {
+ console.error(`[llm-parse] Receipt read failed for "${file.originalName}" (provider=${config.provider}):`, err instanceof Error ? err.message : err);
+ return { receipt: null, warnings: [`${file.originalName}: AI parsing failed — ${err instanceof Error ? err.message : String(err)}`] };
+ }
+
+ const receipt = toReceiptRead(raw);
+ return receipt ? { receipt, warnings: [] } : { receipt: null, warnings: [`${file.originalName}: no receipt could be read`] };
+ }
+
+ private async visionFor(config: ResolvedLlmConfig): Promise {
+ if (config.vision !== 'auto') return config.vision === 'on';
+ // Only a local Ollama can be asked. A cloud model is not assumed to read
+ // images: an admin who knows it does says so.
+ if (config.provider !== 'local') return false;
+ const capabilities = await this.local.modelCapabilities(config.baseUrl, config.model);
+ return capabilities?.includes('vision') ?? false;
+ }
+
async parse(file: { buffer: Buffer; originalName: string }, userId: number): Promise {
const config = this.llmConfig.resolve(userId);
if (!config) return { kiItems: [], warnings: ['AI parsing is not configured'] };
@@ -51,11 +136,22 @@ export class LlmParseService {
apiKey: config.apiKey,
};
- // Native PDF only for Anthropic (its document block reads text AND scans).
- // OpenAI-compatible servers (incl. Ollama/NuExtract) can't ingest PDFs/`file`
- // parts, so every other provider gets extracted text.
+ const imageType = imageMimeType(file.originalName);
+ if (imageType && !(await this.visionFor(config))) {
+ return { kiItems: [], warnings: [`${file.originalName}: the configured model does not read images`] };
+ }
+
+ // A photo goes as an image to every provider. Native PDF only for Anthropic
+ // (its document block reads text AND scans). OpenAI-compatible servers
+ // (incl. Ollama/NuExtract) can't ingest PDFs/`file` parts, so every other
+ // provider gets extracted text.
try {
- if (config.provider === 'anthropic' && isPdf(file.originalName)) {
+ if (imageType) {
+ input.file = await capImage(file.buffer, imageType);
+ console.debug(
+ `[DEBUG] Photo (${input.file.data.length} bytes) sent to ${config.provider}: ${file.originalName}`,
+ );
+ } else if (config.provider === 'anthropic' && isPdf(file.originalName)) {
input.file = { mimeType: MIME_BY_EXT['.pdf'], data: file.buffer };
console.debug(
`[DEBUG] Extracted (native PDF, ${file.buffer.length} bytes) sent to ${config.provider}: ${file.originalName}`,
@@ -78,10 +174,18 @@ export class LlmParseService {
console.debug(`[DEBUG] Extracted text from ${file.originalName} (${input.text.length} chars):\n`, input.text);
}
if (!input.text.trim()) {
- return {
- kiItems: [],
- warnings: [`${file.originalName}: no readable text found (a scanned PDF needs a cloud/vision provider)`],
- };
+ // A scan has no text layer, but a model that reads images can read its pages.
+ const pages = isPdf(file.originalName) && (await this.visionFor(config)) ? await renderPdfPages(file.buffer) : [];
+ if (pages.length === 0) {
+ return {
+ kiItems: [],
+ warnings: [`${file.originalName}: no readable text found (a scanned PDF needs a model that reads images)`],
+ };
+ }
+ input.text = undefined;
+ input.file = pages[0];
+ input.pageImages = pages.slice(1);
+ console.debug(`[DEBUG] Scanned PDF sent to ${config.provider} as ${pages.length} page image(s): ${file.originalName}`);
}
}
} catch (err) {
@@ -96,13 +200,16 @@ export class LlmParseService {
// templates → decompose + grammar-enforced per-reservation extraction → validate
// + repair. Far more reliable on small CPU models than the single-shot path below
// (which stays for cloud providers, whose strong models handle one-shot well).
- if (config.provider === 'local' && input.text) {
+ if (config.provider === 'local' && (input.text || input.file)) {
try {
- const routed = await routeExtraction(input.text, {
+ const ctx = {
baseUrl: config.baseUrl ?? 'http://localhost:11434/v1',
model: config.model,
apiKey: config.apiKey,
- });
+ };
+ const routed = input.file
+ ? await routeImageExtraction([input.file, ...(input.pageImages ?? [])].map((f) => f.data), ctx)
+ : await routeExtraction(input.text ?? '', ctx);
return { kiItems: routed.kiItems, warnings: [...warnings, ...routed.warnings] };
} catch (err) {
console.error(`[llm-parse] AI parsing failed for "${file.originalName}" (provider=${config.provider}):`, err instanceof Error ? err.message : err);
diff --git a/server/src/nest/llm-parse/llm-provider.interface.ts b/server/src/nest/llm-parse/llm-provider.interface.ts
index 945af0cc2..f74096a59 100644
--- a/server/src/nest/llm-parse/llm-provider.interface.ts
+++ b/server/src/nest/llm-parse/llm-provider.interface.ts
@@ -1,4 +1,4 @@
-/** A single binary file (e.g. a PDF) sent natively to a multimodal provider. */
+/** A single binary file (a PDF, or a photo) sent natively to a provider that reads it. */
export interface LlmExtractionFile {
mimeType: string;
data: Buffer;
@@ -15,8 +15,17 @@ export interface LlmExtractionInput {
apiKey?: string;
/** Pre-extracted text (text-like files, or text-only-model mode). */
text?: string;
- /** Native binary (PDF) for multimodal providers. */
+ /** Native binary: a PDF for Anthropic, a photo for a model that reads images. */
file?: LlmExtractionFile;
+ /** The following pages of a scanned PDF, as images after `file`, which holds the first. */
+ pageImages?: LlmExtractionFile[];
+ /**
+ * The key the answer's array sits under, `jsonSchema` being `{ [rootKey]: [...] }`.
+ * Defaults to `reservations`; the receipt read asks for `receipts`.
+ */
+ rootKey?: string;
+ /** The user-turn instruction. Defaults to the booking import's. */
+ userText?: string;
}
/**
diff --git a/server/src/nest/llm-parse/receipt-read.ts b/server/src/nest/llm-parse/receipt-read.ts
new file mode 100644
index 000000000..fefd49901
--- /dev/null
+++ b/server/src/nest/llm-parse/receipt-read.ts
@@ -0,0 +1,108 @@
+import { receiptReadSchema, type ReceiptLine, type ReceiptRead } from '@trek/shared';
+import { toIsoCurrency } from './currency-code';
+import { parseAmount } from './clients/nuextract';
+
+/**
+ * Reading a photographed receipt for the Costs tab: the prompt, the schema the
+ * answer is held to, and the step that turns the answer into a `ReceiptRead`.
+ * Pure, so every piece is testable without a model.
+ *
+ * The field list is flat and short on purpose. A till roll is printed small and
+ * photographed under bad light, and a small local model fills a handful of named
+ * fields far more reliably than the nested schema.org shape bookings use.
+ */
+
+/** Key the cloud clients read the answer's array under (`{ receipts: [...] }`). */
+export const RECEIPT_ROOT_KEY = 'receipts';
+
+export const RECEIPT_USER_TEXT = 'Read the receipt in the attached photo.';
+
+/** One receipt, flat. Ollama compiles it to a grammar, so the answer has this shape or none. */
+export const RECEIPT_JSON_SCHEMA = {
+ type: 'object',
+ properties: {
+ merchant: { type: 'string' },
+ date: { type: 'string' },
+ total: { type: 'number' },
+ currency: { type: 'string' },
+ items: {
+ type: 'array',
+ items: {
+ type: 'object',
+ properties: { name: { type: 'string' }, price: { type: 'number' } },
+ required: ['name', 'price'],
+ },
+ },
+ },
+ // Every field is required. Ollama's grammar lets a small model skip an
+ // optional key, and it does: qwen3.5:4b answered the total and the lines of a
+ // receipt printing both its name and its date, and nothing else. A field the
+ // receipt does not show comes back empty, which toReceiptRead reads as absent.
+ required: ['merchant', 'date', 'total', 'currency', 'items'],
+} as const;
+
+/** The same receipt wrapped in the list the cloud clients' structured output returns. */
+export const RECEIPT_LIST_JSON_SCHEMA = {
+ type: 'object',
+ properties: { [RECEIPT_ROOT_KEY]: { type: 'array', items: RECEIPT_JSON_SCHEMA } },
+ required: [RECEIPT_ROOT_KEY],
+} as const;
+
+export function buildReceiptPrompt(today: Date = new Date()): string {
+ const todayIso = today.toISOString().slice(0, 10);
+ return [
+ 'You read a photographed receipt, bill or invoice: often a paper till roll, possibly crumpled, skewed or in a foreign language.',
+ 'Fill "total" with the grand total actually paid, after tax and tip: the final TOTAL or AMOUNT DUE line, not a subtotal.',
+ 'Fill "currency" with its ISO 4217 code, from the symbol and the country when the code is not printed.',
+ 'Fill "merchant" with the business name printed at the top, and "date" as YYYY-MM-DD.',
+ // A till roll prints "26-08-23" and means one of three dates. The merchant's
+ // country settles day/month order; nothing on the paper settles which pair is
+ // the year, and a receipt is scanned soon after it is paid.
+ `Today is ${todayIso}. When a date could be read several ways, take the reading nearest to today and never one in the future.`,
+ 'Fill "items" with the purchased lines as printed, each with its line total as "price". Leave out subtotal, tax, discount, tip and total lines.',
+ 'Never invent a value: when the receipt does not show a field, answer an empty string for it (0 for "total", [] for "items"). Numbers are plain numbers: 12.5, not "12,50 €".',
+ ].join('\n');
+}
+
+function toText(value: unknown): string | null {
+ return typeof value === 'string' && value.trim() ? value.trim() : null;
+}
+
+/** A real calendar date as YYYY-MM-DD, or null. */
+function toDate(value: unknown): string | null {
+ const s = toText(value);
+ if (!s || !/^\d{4}-\d{2}-\d{2}$/.test(s)) return null;
+ const d = new Date(`${s}T00:00:00Z`);
+ return !Number.isNaN(d.getTime()) && d.toISOString().slice(0, 10) === s ? s : null;
+}
+
+/**
+ * The model's answer as a `ReceiptRead`, or null when it holds nothing a person
+ * could use: no total, no merchant and no line. Each field is kept only when it
+ * reads as what it claims to be, so a garbled one is left for the person to fill
+ * rather than saved wrong.
+ */
+export function toReceiptRead(raw: unknown): ReceiptRead | null {
+ if (!raw || typeof raw !== 'object' || Array.isArray(raw)) return null;
+ const node = raw as Record;
+ const total = parseAmount(node.total);
+ const items: ReceiptLine[] = [];
+ if (Array.isArray(node.items)) {
+ for (const line of node.items) {
+ if (!line || typeof line !== 'object') continue;
+ const name = toText((line as Record).name);
+ const price = parseAmount((line as Record).price);
+ if (name && price !== null) items.push({ name, price });
+ }
+ }
+ const read = {
+ merchant: toText(node.merchant),
+ date: toDate(node.date),
+ total: total !== null && total > 0 ? total : null,
+ currency: toIsoCurrency(node.currency) ?? null,
+ items,
+ };
+ if (read.total === null && read.merchant === null && items.length === 0) return null;
+ const parsed = receiptReadSchema.safeParse(read);
+ return parsed.success ? parsed.data : null;
+}
diff --git a/server/src/nest/llm-parse/router/extraction-router.ts b/server/src/nest/llm-parse/router/extraction-router.ts
index 56dfadf3a..acc8ce841 100644
--- a/server/src/nest/llm-parse/router/extraction-router.ts
+++ b/server/src/nest/llm-parse/router/extraction-router.ts
@@ -23,6 +23,7 @@ import type { KiReservation } from '../../booking-import/kitinerary.types';
import { nuExtractToKiReservations } from '../clients/nuextract';
import { FLAT_SCHEMA_BY_TYPE, FLAT_TYPES, FLIGHTS_ARRAY_SCHEMA, UNION_SINGLE_SCHEMA, type FlatType, type FlatLike } from './flat-schemas';
import { extractEnforced } from './ollama-format.client';
+import { parseLenientJson } from '../lenient-json';
export interface RouterContext {
baseUrl: string;
@@ -45,6 +46,24 @@ const TYPE_HINT: Record = {
event: 'event/attraction. name = the event/ticket, address = the venue, start_time/end_time = full ISO, price/currency = ticket price.',
};
+/**
+ * The union prompt for a photo. Text has keywords that pick a type-specific
+ * schema and hint before the call; a photo has none, so the model gets every
+ * type's field meanings at once. Without them qwen3.5:4b read a train ticket
+ * as bare clock times with no stations, and its price as a JSON object inside
+ * the string.
+ */
+const PHOTO_UNION_HINT = [
+ 'Pick the correct "type", then fill its fields:',
+ '- flight: vehicle_number = flight number, from_code/to_code = IATA codes.',
+ '- train, bus, ferry: from_name/to_name = stations, stops or ports, vehicle_number = train or line number.',
+ '- car: operator = rental company, from_name = pick-up location, to_name = return location.',
+ '- hotel: name, address, checkin_time/checkout_time.',
+ '- restaurant, event: name, address = the venue, start_time (and end_time).',
+ "Every time is full ISO 'YYYY-MM-DDTHH:MM:00' (a transport uses departure_time/arrival_time).",
+ 'price = the total paid as a plain number like 49.00, currency = its ISO 4217 code.',
+].join('\n');
+
/**
* Keyword → reservation type, so an obvious document skips the costlier union/strong path.
*
@@ -213,21 +232,23 @@ async function extractFlights(text: string, ctx: RouterContext): Promise {
+async function extractSingle(text: string, ctx: RouterContext, images?: Buffer[]): Promise {
const known = detectType(text);
const call = (schema: Record, hint: string) =>
extractEnforced({
baseUrl: ctx.baseUrl, model: ctx.model, apiKey: ctx.apiKey,
system: `Extract the single reservation from the document into the flat fields. ${hint} Omit any field that is truly absent.`,
- user: `Document:\n${text}`,
+ user: images?.length ? 'The document is the attached photo.' : `Document:\n${text}`,
schema,
+ // A photo costs far more prompt tokens than the text of the same page.
+ ...(images?.length ? { images: images.map((image) => image.toString('base64')), numCtx: 16384 } : {}),
});
if (known) {
const out = (await call(FLAT_SCHEMA_BY_TYPE[known], `It is a ${TYPE_HINT[known]}`)) ?? {};
return fixArrivalDate(normalizeDates({ ...out, type: known }));
}
- const out = (await call(UNION_SINGLE_SCHEMA, 'Pick the correct "type".')) ?? {};
+ const out = (await call(UNION_SINGLE_SCHEMA, images?.length ? PHOTO_UNION_HINT : 'Pick the correct "type".')) ?? {};
// "I could not tell" used to be written down as "hotel", and a hotel is a
// booking, so a ferry voucher imported from the Transport tab opened the
// booking form with no transport type to pick and stayed 'other' (#2076).
@@ -298,3 +319,38 @@ export async function routeExtraction(text: string, ctx: RouterContext): Promise
const kiItems = nuExtractToKiReservations(flats as unknown as Record[]) as unknown as KiReservation[];
return { kiItems, warnings };
}
+
+/**
+ * The router for a photographed document: one enforced call with the photo
+ * attached, or the pages of a scanned PDF, one image each. There is no text to
+ * spot a flight number or a vendor in, so the model picks the type, and the
+ * booking-wide fields come from its answer alone.
+ */
+export async function routeImageExtraction(images: Buffer[], ctx: RouterContext): Promise<{ kiItems: KiReservation[]; warnings: string[] }> {
+ let flat: FlatLike;
+ try {
+ flat = unwrapPrice(await extractSingle('', ctx, images));
+ } catch (err) {
+ return { kiItems: [], warnings: [`AI parsing failed — ${err instanceof Error ? err.message : String(err)}`] };
+ }
+ const kiItems = nuExtractToKiReservations([flat] as unknown as Record[]) as unknown as KiReservation[];
+ return { kiItems, warnings: [] };
+}
+
+/**
+ * The price as the amount alone. With no document text to take the total from,
+ * the model's `price` is all there is, and qwen3.5:4b writes it as an object
+ * inside the string (`{"amount": 49.00, "currency": "EUR"}`), which the amount
+ * parser then reads as 4900.
+ */
+function unwrapPrice(flat: FlatLike): FlatLike {
+ if (typeof flat.price !== 'string' || !flat.price.trim().startsWith('{')) return flat;
+ const inner = parseLenientJson(flat.price);
+ if (!inner || typeof inner !== 'object' || Array.isArray(inner)) return flat;
+ const { amount, currency } = inner as { amount?: unknown; currency?: unknown };
+ const out: FlatLike = { ...flat };
+ if (typeof amount === 'number' || typeof amount === 'string') out.price = String(amount);
+ else delete out.price;
+ if (!out.currency && typeof currency === 'string') out.currency = currency;
+ return out;
+}
diff --git a/server/src/nest/llm-parse/router/ollama-format.client.ts b/server/src/nest/llm-parse/router/ollama-format.client.ts
index 58e64f36f..d4b226767 100644
--- a/server/src/nest/llm-parse/router/ollama-format.client.ts
+++ b/server/src/nest/llm-parse/router/ollama-format.client.ts
@@ -27,6 +27,8 @@ export interface EnforcedExtractInput {
numPredict?: number;
/** Context window. 8192 fits a typical multi-section booking; raise for long itineraries. */
numCtx?: number;
+ /** Base64 images attached to the user turn, for a vision model reading a photo. */
+ images?: string[];
}
/** Resolve the native API base from a config base URL that may end in `/v1`. */
@@ -59,7 +61,7 @@ export async function extractEnforced(input: EnforcedExtractInput): Promise {
+ const trip = this.db.canAccessTrip(tripId, user.id);
+ if (!trip) throw new HttpException({ error: 'Trip not found' }, 404);
+ if (!this.addons.isAddonEnabled(ADDON_IDS.BUDGET)) throw new HttpException({ error: 'Costs addon is not enabled' }, 404);
+ if (!this.permissions.checkPermission('budget_edit', user.role, trip.user_id, user.id, trip.user_id !== user.id)) {
+ throw new HttpException({ error: 'No permission' }, 403);
+ }
+ if (!file) throw new HttpException({ error: 'No file uploaded' }, 400);
+ if (!imageMimeType(file.originalname)) {
+ throw new HttpException({ error: `Unsupported file type: ${file.originalname}. Accepted: JPG, PNG, WEBP` }, 400);
+ }
+ if (!(await this.llmParse.readsImages(user.id))) {
+ throw new HttpException({ error: 'The configured AI model does not read photos' }, 400);
+ }
+ return { jobId: this.importJobs.startReceipt(tripId, file, user.id) };
+ }
+}
diff --git a/server/src/nest/receipt-scan/receipt-scan.module.ts b/server/src/nest/receipt-scan/receipt-scan.module.ts
new file mode 100644
index 000000000..c768bb43e
--- /dev/null
+++ b/server/src/nest/receipt-scan/receipt-scan.module.ts
@@ -0,0 +1,13 @@
+import { Module } from '@nestjs/common';
+import { ReceiptScanController } from './receipt-scan.controller';
+import { BookingImportModule } from '../booking-import/booking-import.module';
+import { LlmParseModule } from '../llm-parse/llm-parse.module';
+import { AddonsModule } from '../addons/addons.module';
+import { PermissionsModule } from '../permissions/permissions.module';
+
+/** The Costs tab's receipt scan: an HTTP surface over the import job queue. */
+@Module({
+ imports: [BookingImportModule, LlmParseModule, AddonsModule, PermissionsModule],
+ controllers: [ReceiptScanController],
+})
+export class ReceiptScanModule {}
diff --git a/server/src/nest/reservation-import/reservation-import.controller.ts b/server/src/nest/reservation-import/reservation-import.controller.ts
index 25a1c81e7..3874d4df0 100644
--- a/server/src/nest/reservation-import/reservation-import.controller.ts
+++ b/server/src/nest/reservation-import/reservation-import.controller.ts
@@ -27,8 +27,9 @@ import type { AirtrailImportResult } from '@trek/shared';
import { bookingImportModeSchema } from '@trek/shared';
import type { BookingImportPreviewItem, BookingImportPreviewResponse, BookingImportConfirmResponse, BookingImportMode } from '@trek/shared';
import { BookingImportConfirmDto, BookingImportPreviewDto } from './reservation-import.dto';
+import { IMAGE_EXTENSIONS, imageMimeType } from '../llm-parse/image-input';
-const ACCEPTED_EXTS = new Set(['.eml', '.pdf', '.pkpass', '.html', '.htm', '.txt']);
+const ACCEPTED_EXTS = new Set(['.eml', '.pdf', '.pkpass', '.html', '.htm', '.txt', ...IMAGE_EXTENSIONS]);
const MAX_FILE_BYTES = 10 * 1024 * 1024;
const MAX_FILES = 5;
@@ -99,7 +100,7 @@ export class ReservationImportController {
/** Shared validation for both the sync and async import endpoints; returns the parsed mode. */
- private validateImport(tripId: string, user: User, files: Express.Multer.File[] | undefined, rawMode?: string): BookingImportMode {
+ private async validateImport(tripId: string, user: User, files: Express.Multer.File[] | undefined, rawMode?: string): Promise {
const modeResult = bookingImportModeSchema.safeParse(rawMode ?? 'no-ai');
if (!modeResult.success) throw new HttpException({ error: 'Invalid mode' }, 400);
@@ -118,12 +119,19 @@ export class ReservationImportController {
throw new HttpException({ error: `Unsupported file type: ${f.originalname}. Accepted: EML, PDF, PKPass, HTML, TXT` }, 400);
}
}
+ // A photo has no text layer and no structure: only a model that reads images
+ // can do anything with it, so it is refused up front rather than coming back
+ // as an empty preview.
+ if (files.some((f) => imageMimeType(f.originalname)) && (mode === 'no-ai' || !(await this.bookingImport.readsImages(user.id)))) {
+ throw new HttpException({ error: 'The configured AI model does not read photos' }, 400);
+ }
return mode;
}
/**
* POST /api/trips/:tripId/reservations/import/booking
- * Accepts up to 5 booking confirmation files (EML, PDF, PKPass, HTML, TXT).
+ * Accepts up to 5 booking confirmation files (EML, PDF, PKPass, HTML, TXT, and
+ * photos when the AI model reads images).
* Returns a preview list without persisting anything.
*/
@RequirePermission('reservation_edit')
@@ -135,7 +143,7 @@ export class ReservationImportController {
@UploadedFiles() files: Express.Multer.File[] | undefined,
@Body() body: BookingImportPreviewDto,
): Promise {
- const mode = this.validateImport(tripId, user, files, body?.mode);
+ const mode = await this.validateImport(tripId, user, files, body?.mode);
return this.bookingImport.preview(files!, mode, user.id);
}
@@ -155,7 +163,7 @@ export class ReservationImportController {
@UploadedFiles() files: Express.Multer.File[] | undefined,
@Body() body: BookingImportPreviewDto,
): Promise<{ jobId: string }> {
- const mode = this.validateImport(tripId, user, files, body?.mode);
+ const mode = await this.validateImport(tripId, user, files, body?.mode);
const jobId = this.importJobs.start(tripId, files!, mode, user.id);
return { jobId };
}
diff --git a/server/tests/e2e/receipt-scan.e2e.test.ts b/server/tests/e2e/receipt-scan.e2e.test.ts
new file mode 100644
index 000000000..c33ebdd86
--- /dev/null
+++ b/server/tests/e2e/receipt-scan.e2e.test.ts
@@ -0,0 +1,204 @@
+/**
+ * Photos through the AI Parsing addon, end to end: the booking import taking a
+ * photo, the Costs receipt scan and its job, and the capability the pickers ask.
+ *
+ * Real: the guards, multer, LlmParseService with the instance's `vision`
+ * setting, the image cap, the OpenAI-compatible client, the import job queue
+ * and its status route. Replaced: the kitinerary binary (absent) and the one
+ * outbound call, `safeFetchLlm`, which answers as the provider would.
+ */
+import { describe, it, expect, beforeAll, afterAll, beforeEach, vi } from 'vitest';
+import request from 'supertest';
+import cookieParser from 'cookie-parser';
+import type { Server } from 'http';
+import { Test } from '@nestjs/testing';
+import { Jimp } from 'jimp';
+import { sessionCookie } from './harness';
+
+const { canAccessTrip, safeFetchLlm } = vi.hoisted(() => ({ canAccessTrip: vi.fn(), safeFetchLlm: vi.fn() }));
+vi.mock('../../src/db/database', async () => {
+ const { default: Database } = await import('better-sqlite3');
+ const tmp = new Database(':memory:');
+ tmp.exec('PRAGMA journal_mode = WAL');
+ tmp.exec('PRAGMA foreign_keys = ON');
+ return { db: tmp, canAccessTrip, isOwner: vi.fn(() => true), getPlaceWithTags: vi.fn(), closeDb: () => {}, reinitialize: () => {} };
+});
+vi.mock('../../src/websocket', () => ({ broadcast: vi.fn(), broadcastToUser: vi.fn() }));
+vi.mock('../../src/utils/ssrfGuard', async (orig) => ({ ...(await orig>()), safeFetchLlm }));
+
+import { db } from '../../src/db/database';
+import { createTables } from '../../src/db/schema';
+import { runMigrations } from '../../src/db/migrations';
+import { DatabaseModule } from '../../src/nest/database/database.module';
+import { RealtimeModule } from '../../src/nest/realtime/realtime.module';
+import { ReservationImportModule } from '../../src/nest/reservation-import/reservation-import.module';
+import { ReceiptScanModule } from '../../src/nest/receipt-scan/receipt-scan.module';
+import { LlmParseModule } from '../../src/nest/llm-parse/llm-parse.module';
+import { KitineraryExtractorService } from '../../src/nest/booking-import/kitinerary-extractor.service';
+import { NotificationsService } from '../../src/nest/notifications/notifications.service';
+import { PermissionsService } from '../../src/nest/permissions/permissions.service';
+import { TrekExceptionFilter } from '../../src/nest/common/trek-exception.filter';
+import { ZodValidationPipe } from '../../src/nest/common/zod-validation.pipe';
+
+function setVision(vision: string) {
+ db.prepare("UPDATE addons SET config = ? WHERE id = 'llm_parsing'").run(
+ JSON.stringify({ provider: 'openai', model: 'gpt-4.1-mini', apiKey: '', vision }),
+ );
+}
+
+/** The provider's chat-completions answer, carrying `payload` as its content. */
+function providerAnswers(payload: unknown) {
+ safeFetchLlm.mockResolvedValue({
+ ok: true,
+ status: 200,
+ json: async () => ({ choices: [{ message: { content: JSON.stringify(payload) } }] }),
+ text: async () => '',
+ } as unknown as Response);
+}
+
+const sentUserContent = () => JSON.parse((safeFetchLlm.mock.calls[0][1] as RequestInit).body as string).messages[1].content;
+
+describe('Photos through AI Parsing e2e', () => {
+ let server: Server;
+ let app: Awaited>;
+ let tripId: number;
+ let photo: Buffer;
+
+ async function build() {
+ const moduleRef = await Test.createTestingModule({ imports: [DatabaseModule, RealtimeModule, LlmParseModule, ReservationImportModule, ReceiptScanModule] })
+ .overrideProvider(KitineraryExtractorService)
+ .useValue({ onModuleInit: () => {}, isAvailable: () => false, extract: vi.fn(), describe: () => ({ available: false }) })
+ .overrideProvider(NotificationsService)
+ .useValue({ send: vi.fn().mockResolvedValue(undefined) })
+ .compile();
+ const nest = moduleRef.createNestApplication();
+ nest.use(cookieParser());
+ nest.useGlobalFilters(new TrekExceptionFilter());
+ nest.useGlobalPipes(new ZodValidationPipe());
+ await nest.init();
+ return nest;
+ }
+
+ beforeAll(async () => {
+ createTables(db);
+ runMigrations(db);
+ db.prepare(
+ "INSERT INTO users (id, username, email, password_hash, role, password_version) VALUES (1, 'e2e-user', 'e2e@example.test', 'x', 'user', 0)",
+ ).run();
+ tripId = Number(db.prepare("INSERT INTO trips (user_id, title) VALUES (1, 'Lyon')").run().lastInsertRowid);
+ db.prepare(
+ `INSERT INTO addons (id, name, type, enabled, config) VALUES ('llm_parsing', 'AI Parsing', 'integration', 1, '{}')
+ ON CONFLICT(id) DO UPDATE SET enabled = 1`,
+ ).run();
+ db.prepare(
+ `INSERT INTO addons (id, name, type, enabled) VALUES ('budget', 'Costs', 'trip', 1)
+ ON CONFLICT(id) DO UPDATE SET enabled = 1`,
+ ).run();
+ photo = Buffer.from(await new Jimp({ width: 40, height: 60, color: 0xffffffff }).getBuffer('image/png'));
+ app = await build();
+ vi.spyOn(app.get(PermissionsService), 'checkPermission').mockReturnValue(true);
+ server = app.getHttpServer();
+ });
+
+ beforeEach(() => {
+ canAccessTrip.mockImplementation((id: unknown) => db.prepare('SELECT * FROM trips WHERE id = ?').get(id));
+ safeFetchLlm.mockReset();
+ setVision('on');
+ });
+
+ afterAll(async () => {
+ await app.close();
+ });
+
+ describe('GET /api/llm/capabilities', () => {
+ it('401 without a cookie', async () => {
+ expect((await request(server).get('/api/llm/capabilities')).status).toBe(401);
+ });
+
+ it('answers the instance setting, and no for a cloud model on Automatic', async () => {
+ expect((await request(server).get('/api/llm/capabilities').set('Cookie', sessionCookie(1))).body).toEqual({ images: true });
+ setVision('auto');
+ expect((await request(server).get('/api/llm/capabilities').set('Cookie', sessionCookie(1))).body).toEqual({ images: false });
+ expect(safeFetchLlm).not.toHaveBeenCalled();
+ });
+ });
+
+ describe('booking import', () => {
+ const upload = () =>
+ request(server)
+ .post(`/api/trips/${tripId}/reservations/import/booking`)
+ .set('Cookie', sessionCookie(1))
+ .field('mode', 'fallback-on-empty')
+ .attach('files', photo, { filename: 'ticket.png', contentType: 'image/png' });
+
+ it('sends a photo to the model as an image, and maps its answer', async () => {
+ providerAnswers({ reservations: [{ '@type': 'TrainReservation', reservationFor: { departureStation: { name: 'Lyon Part-Dieu' }, arrivalStation: { name: 'Paris Gare de Lyon' }, departureTime: '2026-10-03T08:04:00' } }] });
+ const res = await upload();
+
+ expect(res.status).toBe(201);
+ expect(res.body.items).toHaveLength(1);
+ expect(res.body.files).toEqual([{ fileName: 'ticket.png', aiAvailable: true, aiUsed: true }]);
+ const image = sentUserContent().find((p: { type: string }) => p.type === 'image_url');
+ expect(image.image_url.url).toMatch(/^data:image\/png;base64,/);
+ });
+
+ it('refuses a photo with 400 when the model reads no images', async () => {
+ setVision('off');
+ const res = await upload();
+ expect(res.status).toBe(400);
+ expect(res.body.error).toBe('The configured AI model does not read photos');
+ expect(safeFetchLlm).not.toHaveBeenCalled();
+ });
+ });
+
+ describe('POST /api/trips/:tripId/budget/receipt-scan', () => {
+ const scan = (name = 'bill.png') =>
+ request(server)
+ .post(`/api/trips/${tripId}/budget/receipt-scan`)
+ .set('Cookie', sessionCookie(1))
+ .attach('file', photo, { filename: name, contentType: name.endsWith('.png') ? 'image/png' : 'application/pdf' });
+
+ it('401 without a cookie, 404 for a trip out of reach', async () => {
+ expect((await request(server).post(`/api/trips/${tripId}/budget/receipt-scan`)).status).toBe(401);
+ canAccessTrip.mockReturnValue(undefined);
+ expect((await scan()).status).toBe(404);
+ });
+
+ it('reads the receipt in a job whose status answers what was read', async () => {
+ providerAnswers({ receipts: [{ merchant: 'Boulangerie du Port', date: '2026-09-21', total: 18.1, currency: 'EUR', items: [{ name: 'Quiche', price: 6.9 }] }] });
+ const res = await scan();
+ expect(res.status).toBe(201);
+ const { jobId } = res.body;
+
+ let status: request.Response | undefined;
+ await vi.waitFor(async () => {
+ status = await request(server).get(`/api/trips/${tripId}/reservations/import/jobs/${jobId}`).set('Cookie', sessionCookie(1));
+ expect(status.body.status).toBe('done');
+ });
+ expect(status!.body.result).toEqual({
+ receipt: { merchant: 'Boulangerie du Port', date: '2026-09-21', total: 18.1, currency: 'EUR', items: [{ name: 'Quiche', price: 6.9 }] },
+ warnings: [],
+ });
+ const body = JSON.parse((safeFetchLlm.mock.calls[0][1] as RequestInit).body as string);
+ expect(body.response_format.json_schema.name).toBe('receipts');
+ });
+
+ it('refuses a file that is not a photo, and any photo when the model reads no images', async () => {
+ expect((await scan('bill.pdf')).status).toBe(400);
+ setVision('off');
+ expect((await scan()).status).toBe(400);
+ expect(safeFetchLlm).not.toHaveBeenCalled();
+ });
+
+ it('404 while Costs is off', async () => {
+ db.prepare("UPDATE addons SET enabled = 0 WHERE id = 'budget'").run();
+ try {
+ const res = await scan();
+ expect(res.status).toBe(404);
+ expect(res.body.error).toBe('Costs addon is not enabled');
+ } finally {
+ db.prepare("UPDATE addons SET enabled = 1 WHERE id = 'budget'").run();
+ }
+ });
+ });
+});
diff --git a/server/tests/unit/nest/booking-import/booking-import.service.test.ts b/server/tests/unit/nest/booking-import/booking-import.service.test.ts
index 8226db8a7..8859003d1 100644
--- a/server/tests/unit/nest/booking-import/booking-import.service.test.ts
+++ b/server/tests/unit/nest/booking-import/booking-import.service.test.ts
@@ -79,6 +79,18 @@ describe('BookingImportService.preview', () => {
expect(res.files![0].aiUsed).toBe(false);
});
+ it('fallback-on-empty: a photo goes straight to the LLM, kitinerary is not asked', async () => {
+ const { svc, extractor, llmParse } = make({
+ kit: true, ai: true,
+ extract: async () => [HOTEL_KI],
+ parse: async () => ({ kiItems: [HOTEL_KI], warnings: [] }),
+ });
+ const res = await svc.preview([file('ticket.jpg')], 'fallback-on-empty', 1);
+ expect(extractor.extract).not.toHaveBeenCalled();
+ expect(llmParse.parse).toHaveBeenCalledWith({ buffer: expect.any(Buffer), originalName: 'ticket.jpg' }, 1);
+ expect(res.items[0].needs_review).toBe(true);
+ });
+
it('force-ai: skips kitinerary entirely and uses the LLM', async () => {
const { svc, extractor, llmParse } = make({
kit: true, ai: true,
diff --git a/server/tests/unit/nest/booking-import/import-jobs.service.test.ts b/server/tests/unit/nest/booking-import/import-jobs.service.test.ts
index da170760a..ece990910 100644
--- a/server/tests/unit/nest/booking-import/import-jobs.service.test.ts
+++ b/server/tests/unit/nest/booking-import/import-jobs.service.test.ts
@@ -7,8 +7,8 @@ import { ImportJobsService } from '../../../../src/nest/booking-import/import-jo
import { RealtimeService } from '../../../../src/nest/realtime/realtime.service';
type Preview = ReturnType;
-function makeService(preview: Preview) {
- return new ImportJobsService({ preview } as never, new RealtimeService());
+function makeService(preview: Preview, readReceipt: Preview = vi.fn()) {
+ return new ImportJobsService({ preview } as never, new RealtimeService(), { readReceipt } as never);
}
const files = (n: number) => Array.from({ length: n }, (_, i) => ({ originalname: `f${i}.pdf` })) as never;
const eventsFor = (jobId: string) => broadcastToUser.mock.calls.map((c) => c[1]).filter((p) => p.jobId === jobId);
@@ -101,3 +101,30 @@ describe('ImportJobsService', () => {
expect(svc.get(id, 99)).toBeUndefined();
});
});
+
+describe('ImportJobsService.startReceipt', () => {
+ it('reads the receipt off-request and pushes the read as the result', async () => {
+ const result = { receipt: { merchant: 'Shop', date: null, total: 4, currency: 'EUR', items: [] }, warnings: [] };
+ const readReceipt = vi.fn(async () => result);
+ const svc = makeService(vi.fn(), readReceipt);
+ const id = svc.startReceipt('7', { originalname: 'r.jpg', buffer: Buffer.from('x') } as never, 42);
+
+ await vi.waitFor(() => expect(svc.get(id, 42)?.status).toBe('done'));
+ expect(svc.get(id, 42)).toMatchObject({ result, done: 1, total: 1 });
+ expect(readReceipt).toHaveBeenCalledWith({ buffer: Buffer.from('x'), originalName: 'r.jpg' }, 42);
+ expect(eventsFor(id).find((p) => p.type === 'import:done')).toMatchObject({ result, tripId: '7' });
+ });
+
+ it('queues behind the same user\'s booking parse rather than running beside it', async () => {
+ let release!: () => void;
+ const preview = vi.fn(() => new Promise((resolve) => { release = () => resolve({ items: [] }); }));
+ const readReceipt = vi.fn(async () => ({ receipt: null, warnings: [] }));
+ const svc = makeService(preview, readReceipt);
+ svc.start('7', files(1), 'force-ai', 42);
+ const id = svc.startReceipt('7', { originalname: 'r.jpg', buffer: Buffer.from('x') } as never, 42);
+ await new Promise((r) => setTimeout(r, 10));
+ expect(readReceipt).not.toHaveBeenCalled();
+ release();
+ await vi.waitFor(() => expect(svc.get(id, 42)?.status).toBe('done'));
+ });
+});
diff --git a/server/tests/unit/nest/booking-import/reservation-import.controller.test.ts b/server/tests/unit/nest/booking-import/reservation-import.controller.test.ts
index cf51572ce..87ea2202b 100644
--- a/server/tests/unit/nest/booking-import/reservation-import.controller.test.ts
+++ b/server/tests/unit/nest/booking-import/reservation-import.controller.test.ts
@@ -14,6 +14,7 @@ function make(over: Partial = {}) {
canEdit: vi.fn(() => true),
isAvailable: vi.fn(() => true),
aiAvailable: vi.fn(() => true),
+ readsImages: vi.fn(async () => true),
preview: vi.fn(async () => ({ items: [], warnings: [], files: [] })),
...over,
} as unknown as BookingImportService;
@@ -55,6 +56,28 @@ describe('ReservationImportController.preview', () => {
expect(svc.preview).toHaveBeenCalledWith([expect.anything()], 'fallback-on-empty', 1);
});
+ it('takes a photo when the model reads images, without asking otherwise', async () => {
+ const { c, svc } = make();
+ await c.preview(user, 't1', [file('ticket.JPG')], { mode: 'fallback-on-empty' });
+ expect(svc.preview).toHaveBeenCalledWith([expect.anything()], 'fallback-on-empty', 1);
+ (svc.readsImages as ReturnType).mockClear();
+ await c.preview(user, 't1', [file('a.pdf')], { mode: 'fallback-on-empty' });
+ expect(svc.readsImages).not.toHaveBeenCalled();
+ });
+
+ it('refuses a photo with 400 when the model does not read images, or AI is not asked', async () => {
+ const { c, svc } = make({ readsImages: vi.fn(async () => false) });
+ expect(await status(() => c.preview(user, 't1', [file('a.pdf'), file('ticket.png')], { mode: 'fallback-on-empty' }))).toBe(400);
+ const { c: c2 } = make();
+ expect(await status(() => c2.preview(user, 't1', [file('ticket.png')], { mode: 'no-ai' }))).toBe(400);
+ expect(svc.preview).not.toHaveBeenCalled();
+ });
+
+ it('still refuses a file type it does not know, HEIC included', async () => {
+ const { c } = make();
+ expect(await status(() => c.preview(user, 't1', [file('IMG_1.heic')], { mode: 'fallback-on-empty' }))).toBe(400);
+ });
+
it('defaults the mode to no-ai when omitted', async () => {
const { c, svc } = make();
await c.preview(user, 't1', [file()], {});
diff --git a/server/tests/unit/nest/llm-parse/clients.test.ts b/server/tests/unit/nest/llm-parse/clients.test.ts
index 9a151dbeb..aed940545 100644
--- a/server/tests/unit/nest/llm-parse/clients.test.ts
+++ b/server/tests/unit/nest/llm-parse/clients.test.ts
@@ -275,6 +275,15 @@ describe('OpenAiCompatibleClient', () => {
expect(safeFetchLlmMock).toHaveBeenCalledTimes(1);
});
+ it('sends each page of a scanned PDF as its own image_url, in order', async () => {
+ const fetchFn = mockFetch(() => jsonResponse({ choices: [{ message: { content: '{"reservations":[]}' } }] }));
+ const page = (n: string) => ({ mimeType: 'image/png', data: Buffer.from(n) });
+ await new OpenAiCompatibleClient().extract({ ...baseInput, text: undefined, file: page('p1'), pageImages: [page('p2')] });
+ const parts = JSON.parse((fetchFn.mock.calls[0][1] as RequestInit).body as string).messages[1].content;
+ const urls = parts.filter((p: { type: string }) => p.type === 'image_url').map((p: { image_url: { url: string } }) => p.image_url.url);
+ expect(urls).toEqual([`data:image/png;base64,${Buffer.from('p1').toString('base64')}`, `data:image/png;base64,${Buffer.from('p2').toString('base64')}`]);
+ });
+
it('sends an image natively as image_url but never a file/pdf part', async () => {
const fetchFn = mockFetch(() => jsonResponse({ choices: [{ message: { content: '{"reservations":[]}' } }] }));
await new OpenAiCompatibleClient().extract({ ...baseInput, file: { mimeType: 'image/png', data: Buffer.from('IMG') } });
@@ -380,6 +389,28 @@ describe('OpenAiCompatibleClient — NuExtract path', () => {
});
describe('AnthropicClient', () => {
+ it('sends the pages of a scanned PDF as image blocks after the first', async () => {
+ const fetchFn = mockFetch(() =>
+ jsonResponse({ stop_reason: 'tool_use', content: [{ type: 'tool_use', name: 'emit_reservations', input: { reservations: [] } }] }),
+ );
+ const page = (n: string) => ({ mimeType: 'image/png', data: Buffer.from(n) });
+ await new AnthropicClient().extract({ ...baseInput, text: undefined, file: page('p1'), pageImages: [page('p2')] });
+ const content = JSON.parse((fetchFn.mock.calls[0][1] as RequestInit).body as string).messages[0].content;
+ expect(content.map((b: { type: string }) => b.type)).toEqual(['image', 'image', 'text']);
+ expect(content[1].source.data).toBe(Buffer.from('p2').toString('base64'));
+ });
+
+ it('sends a photo as an image block and a PDF as a document block', async () => {
+ const fetchFn = mockFetch(() =>
+ jsonResponse({ stop_reason: 'tool_use', content: [{ type: 'tool_use', name: 'emit_reservations', input: { reservations: [] } }] }),
+ );
+ await new AnthropicClient().extract({ ...baseInput, text: undefined, file: { mimeType: 'image/jpeg', data: Buffer.from('jpg') } });
+ await new AnthropicClient().extract({ ...baseInput, text: undefined, file: { mimeType: 'application/pdf', data: Buffer.from('pdf') } });
+ const blockOf = (i: number) => JSON.parse((fetchFn.mock.calls[i][1] as RequestInit).body as string).messages[0].content[0];
+ expect(blockOf(0)).toEqual({ type: 'image', source: { type: 'base64', media_type: 'image/jpeg', data: Buffer.from('jpg').toString('base64') } });
+ expect(blockOf(1).type).toBe('document');
+ });
+
it('forces the emit_reservations tool and reads its input', async () => {
const fetchFn = mockFetch(() =>
jsonResponse({ stop_reason: 'tool_use', content: [{ type: 'tool_use', name: 'emit_reservations', input: { reservations: [{ '@type': 'LodgingReservation' }] } }] }),
diff --git a/server/tests/unit/nest/llm-parse/extraction-router.test.ts b/server/tests/unit/nest/llm-parse/extraction-router.test.ts
index cc6ad55e7..cf9972a94 100644
--- a/server/tests/unit/nest/llm-parse/extraction-router.test.ts
+++ b/server/tests/unit/nest/llm-parse/extraction-router.test.ts
@@ -15,6 +15,7 @@ import {
detectFlightNumbers,
fixArrivalDate,
routeExtraction,
+ routeImageExtraction,
} from '../../../../src/nest/llm-parse/router/extraction-router';
const CTX = { baseUrl: 'http://ollama:11434/v1', model: 'qwen3:8b' };
@@ -257,3 +258,64 @@ describe('printed 12-hour clocks and unreadable types (#2094, #2076)', () => {
expect(flats[0].type_guessed).toBeUndefined();
});
});
+
+describe('routeImageExtraction', () => {
+ it('makes one enforced call with the photo attached and lets the model pick the type', async () => {
+ extractEnforced.mockResolvedValue({ type: 'train', from_name: 'Lyon', to_name: 'Paris', price: '42', currency: 'EUR' });
+ const out = await routeImageExtraction([Buffer.from('photo')], CTX);
+ expect(out).toEqual({ kiItems: [{ '@type': 'Mock' }], warnings: [] });
+ expect(extractEnforced).toHaveBeenCalledTimes(1);
+ const call = extractEnforced.mock.calls[0][0];
+ expect(call.images).toEqual([Buffer.from('photo').toString('base64')]);
+ expect(call.numCtx).toBe(16384);
+ expect(call.user).not.toMatch(/Document:/);
+ expect(mapToKi.mock.calls[0][0][0]).toMatchObject({ type: 'train', from_name: 'Lyon', price: '42', currency: 'EUR' });
+ });
+
+ it('tells the model what each type\'s fields mean, since a photo has no keyword to pick a schema', async () => {
+ extractEnforced.mockResolvedValue({ type: 'train' });
+ await routeImageExtraction([Buffer.from('photo')], CTX);
+ const system = extractEnforced.mock.calls[0][0].system;
+ expect(system).toMatch(/from_name\/to_name = stations/);
+ expect(system).toMatch(/full ISO/);
+ });
+
+ it('reads a price the model wrapped as an object inside the string', async () => {
+ extractEnforced.mockResolvedValue({ type: 'train', price: '{"amount": 49.00, "currency": "EUR"}' });
+ await routeImageExtraction([Buffer.from('photo')], CTX);
+ expect(mapToKi.mock.calls[0][0][0]).toMatchObject({ price: '49', currency: 'EUR' });
+
+ mapToKi.mockClear();
+ extractEnforced.mockResolvedValue({ type: 'train', price: '{"amount": "12,50"}', currency: 'CHF' });
+ await routeImageExtraction([Buffer.from('photo')], CTX);
+ expect(mapToKi.mock.calls[0][0][0]).toMatchObject({ price: '12,50', currency: 'CHF' });
+
+ mapToKi.mockClear();
+ extractEnforced.mockResolvedValue({ type: 'train', price: '{"note": "see below"}' });
+ await routeImageExtraction([Buffer.from('photo')], CTX);
+ expect(mapToKi.mock.calls[0][0][0]).not.toHaveProperty('price');
+ });
+
+ it('leaves a plain price, or one that only looks like an object, as it came', async () => {
+ for (const price of ['49,00 EUR', '{not json']) {
+ mapToKi.mockClear();
+ extractEnforced.mockResolvedValue({ type: 'train', price });
+ await routeImageExtraction([Buffer.from('photo')], CTX);
+ expect(mapToKi.mock.calls[0][0][0].price).toBe(price);
+ }
+ });
+
+ it('attaches every page of a scanned PDF to the one call', async () => {
+ extractEnforced.mockResolvedValue({ type: 'hotel' });
+ await routeImageExtraction([Buffer.from('p1'), Buffer.from('p2')], CTX);
+ expect(extractEnforced).toHaveBeenCalledTimes(1);
+ expect(extractEnforced.mock.calls[0][0].images).toEqual([Buffer.from('p1').toString('base64'), Buffer.from('p2').toString('base64')]);
+ });
+
+ it('degrades to a warning when the call throws', async () => {
+ extractEnforced.mockRejectedValue(new Error('timeout'));
+ const out = await routeImageExtraction([Buffer.from('photo')], CTX);
+ expect(out.kiItems).toEqual([]);
+ expect(out.warnings[0]).toMatch(/timeout/);
+ });
+});
diff --git a/server/tests/unit/nest/llm-parse/image-input.test.ts b/server/tests/unit/nest/llm-parse/image-input.test.ts
new file mode 100644
index 000000000..493aba20a
--- /dev/null
+++ b/server/tests/unit/nest/llm-parse/image-input.test.ts
@@ -0,0 +1,94 @@
+import { describe, it, expect } from 'vitest';
+import { Jimp } from 'jimp';
+import { capImage, imageMimeType, IMAGE_MAX_EDGE, renderPdfPages } from '../../../../src/nest/llm-parse/image-input';
+
+async function png(w: number, h: number): Promise {
+ const image = new Jimp({ width: w, height: h, color: 0xffffffff });
+ return Buffer.from(await image.getBuffer('image/png'));
+}
+
+describe('imageMimeType', () => {
+ it('knows the photo formats a provider reads, whatever the case', () => {
+ expect(imageMimeType('a.jpg')).toBe('image/jpeg');
+ expect(imageMimeType('A.JPEG')).toBe('image/jpeg');
+ expect(imageMimeType('a.png')).toBe('image/png');
+ expect(imageMimeType('a.webp')).toBe('image/webp');
+ });
+
+ it('is null for everything else, HEIC included', () => {
+ for (const name of ['a.heic', 'a.pdf', 'a.txt', 'jpg', 'a.jpg.pdf']) expect(imageMimeType(name)).toBeNull();
+ });
+});
+
+describe('capImage', () => {
+ it('shrinks the long edge of a large photo to the cap, as a JPEG', async () => {
+ const out = await capImage(await png(IMAGE_MAX_EDGE * 2, 1000), 'image/png');
+ expect(out.mimeType).toBe('image/jpeg');
+ const back = await Jimp.read(out.data);
+ expect(back.width).toBe(IMAGE_MAX_EDGE);
+ expect(back.height).toBe(500);
+ });
+
+ it('caps a portrait photo by its height', async () => {
+ const back = await Jimp.read((await capImage(await png(800, IMAGE_MAX_EDGE * 2), 'image/png')).data);
+ expect(back.height).toBe(IMAGE_MAX_EDGE);
+ expect(back.width).toBe(400);
+ });
+
+ it('leaves a photo that is already small enough untouched', async () => {
+ const data = await png(400, 300);
+ expect(await capImage(data, 'image/png')).toEqual({ data, mimeType: 'image/png' });
+ });
+
+ it('passes bytes it cannot decode through unchanged', async () => {
+ const data = Buffer.from('RIFF....WEBP');
+ expect(await capImage(data, 'image/webp')).toEqual({ data, mimeType: 'image/webp' });
+ });
+});
+
+/** A PDF with no text layer: each page is one JPEG, as a scanner writes it. */
+async function scannedPdf(pages: number): Promise {
+ const jpg = Buffer.from(await new Jimp({ width: 300, height: 400, color: 0xddddddff }).getBuffer('image/jpeg'));
+ const objects: Buffer[] = [];
+ const kids = Array.from({ length: pages }, (_, i) => `${3 + i * 3} 0 R`).join(' ');
+ objects.push(Buffer.from('<< /Type /Catalog /Pages 2 0 R >>'));
+ objects.push(Buffer.from(`<< /Type /Pages /Kids [${kids}] /Count ${pages} >>`));
+ for (let i = 0; i < pages; i++) {
+ const page = 3 + i * 3;
+ const content = 'q 300 0 0 400 0 0 cm /Im0 Do Q';
+ objects.push(Buffer.from(`<< /Type /Page /Parent 2 0 R /MediaBox [0 0 300 400] /Resources << /XObject << /Im0 ${page + 2} 0 R >> >> /Contents ${page + 1} 0 R >>`));
+ objects.push(Buffer.from(`<< /Length ${content.length} >>\nstream\n${content}\nendstream`));
+ objects.push(Buffer.concat([
+ Buffer.from(`<< /Type /XObject /Subtype /Image /Width 300 /Height 400 /ColorSpace /DeviceRGB /BitsPerComponent 8 /Filter /DCTDecode /Length ${jpg.length} >>\nstream\n`),
+ jpg,
+ Buffer.from('\nendstream'),
+ ]));
+ }
+ const parts: Buffer[] = [Buffer.from('%PDF-1.4\n')];
+ const offsets: number[] = [];
+ let length = parts[0].length;
+ objects.forEach((body, i) => {
+ offsets.push(length);
+ const obj = Buffer.concat([Buffer.from(`${i + 1} 0 obj\n`), body, Buffer.from('\nendobj\n')]);
+ parts.push(obj);
+ length += obj.length;
+ });
+ const rows = offsets.map((o) => `${String(o).padStart(10, '0')} 00000 n \n`).join('');
+ parts.push(Buffer.from(`xref\n0 ${objects.length + 1}\n0000000000 65535 f \n${rows}trailer\n<< /Size ${objects.length + 1} /Root 1 0 R >>\nstartxref\n${length}\n%%EOF\n`));
+ return Buffer.concat(parts);
+}
+
+describe('renderPdfPages', () => {
+ it('draws the first two pages of a scan as images no larger than a capped photo', async () => {
+ const pages = await renderPdfPages(await scannedPdf(3));
+ expect(pages).toHaveLength(2);
+ for (const page of pages) {
+ const image = await Jimp.read(page.data);
+ expect(Math.max(image.width, image.height)).toBeLessThanOrEqual(IMAGE_MAX_EDGE);
+ }
+ }, 30_000);
+
+ it('throws for bytes that are not a PDF, which the caller reports as unreadable', async () => {
+ await expect(renderPdfPages(Buffer.from('not a pdf'))).rejects.toThrow();
+ });
+});
diff --git a/server/tests/unit/nest/llm-parse/llm-capabilities.controller.test.ts b/server/tests/unit/nest/llm-parse/llm-capabilities.controller.test.ts
new file mode 100644
index 000000000..238313c9c
--- /dev/null
+++ b/server/tests/unit/nest/llm-parse/llm-capabilities.controller.test.ts
@@ -0,0 +1,20 @@
+import { describe, it, expect, vi } from 'vitest';
+import { GUARDS_METADATA } from '@nestjs/common/constants';
+import { LlmCapabilitiesController } from '../../../../src/nest/llm-parse/llm-capabilities.controller';
+import { JwtAuthGuard } from '../../../../src/nest/auth/jwt-auth.guard';
+import type { LlmParseService } from '../../../../src/nest/llm-parse/llm-parse.service';
+import type { User } from '../../../../src/types';
+
+describe('LlmCapabilitiesController', () => {
+ it("answers whether the caller's model reads images", async () => {
+ const readsImages = vi.fn(async (id: number) => id === 1);
+ const c = new LlmCapabilitiesController({ readsImages } as unknown as LlmParseService);
+ await expect(c.capabilities({ id: 1 } as User)).resolves.toEqual({ images: true });
+ await expect(c.capabilities({ id: 2 } as User)).resolves.toEqual({ images: false });
+ expect(readsImages).toHaveBeenCalledWith(2);
+ });
+
+ it('needs a session', () => {
+ expect(Reflect.getMetadata(GUARDS_METADATA, LlmCapabilitiesController)).toContain(JwtAuthGuard);
+ });
+});
diff --git a/server/tests/unit/nest/llm-parse/llm-config.resolver.test.ts b/server/tests/unit/nest/llm-parse/llm-config.resolver.test.ts
index 8371e56f4..8319008ab 100644
--- a/server/tests/unit/nest/llm-parse/llm-config.resolver.test.ts
+++ b/server/tests/unit/nest/llm-parse/llm-config.resolver.test.ts
@@ -64,16 +64,25 @@ describe('resolveLlmConfig', () => {
});
it('uses instance config when present (and decrypts the key)', () => {
- setInstanceConfig({ provider: 'anthropic', model: 'claude-opus-4-8', apiKey: 'sk-plain', multimodal: true });
+ setInstanceConfig({ provider: 'anthropic', model: 'claude-opus-4-8', apiKey: 'sk-plain', vision: 'on' });
expect(resolver.resolve(1)).toEqual({
provider: 'anthropic',
model: 'claude-opus-4-8',
baseUrl: undefined,
apiKey: 'sk-plain',
- multimodal: true,
+ vision: 'on',
});
});
+ it('reads the instance vision setting as auto when it is missing, unknown, or only the old multimodal flag', () => {
+ for (const extra of [{}, { vision: 'maybe' }, { multimodal: false }, { multimodal: true }]) {
+ setInstanceConfig({ provider: 'local', model: 'qwen3.5:4b', ...extra });
+ expect(resolver.resolve(1)?.vision).toBe('auto');
+ }
+ setInstanceConfig({ provider: 'local', model: 'qwen3.5:4b', vision: 'off' });
+ expect(resolver.resolve(1)?.vision).toBe('off');
+ });
+
it('instance config with a base URL still wins for a plain user (#1772 does not touch it)', () => {
setInstanceConfig({ provider: 'local', model: 'nuextract', baseUrl: 'http://ollama:11434' });
setRole('user');
@@ -89,7 +98,7 @@ describe('resolveLlmConfig', () => {
model: 'claude-sonnet',
baseUrl: undefined,
apiKey: 'user-key',
- multimodal: true,
+ vision: 'on',
});
expect(getDecryptedUserSetting).toHaveBeenCalledWith(7, 'llm_api_key');
});
@@ -114,7 +123,7 @@ describe('resolveLlmConfig', () => {
model: 'gpt-4o-mini',
baseUrl: undefined,
apiKey: 'sk-user',
- multimodal: false,
+ vision: 'off',
});
});
diff --git a/server/tests/unit/nest/llm-parse/llm-config.test.ts b/server/tests/unit/nest/llm-parse/llm-config.test.ts
new file mode 100644
index 000000000..3a48dd1bc
--- /dev/null
+++ b/server/tests/unit/nest/llm-parse/llm-config.test.ts
@@ -0,0 +1,19 @@
+import { describe, it, expect } from 'vitest';
+import { prepareLlmAddonConfigForWrite } from '../../../../src/nest/llm-parse/llm-config';
+
+describe('prepareLlmAddonConfigForWrite: the vision setting', () => {
+ it('stores a known mode as sent', () => {
+ for (const vision of ['auto', 'on', 'off']) {
+ expect(prepareLlmAddonConfigForWrite({ provider: 'local', vision }, undefined).vision).toBe(vision);
+ }
+ });
+
+ it('stores anything else as auto, so the resolver never meets a value it cannot read', () => {
+ expect(prepareLlmAddonConfigForWrite({ vision: 'maybe' }, undefined).vision).toBe('auto');
+ expect(prepareLlmAddonConfigForWrite({ vision: true }, undefined).vision).toBe('auto');
+ });
+
+ it('adds no vision key to a config that sent none', () => {
+ expect(prepareLlmAddonConfigForWrite({ provider: 'openai', model: 'm' }, undefined)).not.toHaveProperty('vision');
+ });
+});
diff --git a/server/tests/unit/nest/llm-parse/llm-local.service.test.ts b/server/tests/unit/nest/llm-parse/llm-local.service.test.ts
index 76b335de6..a26dd5786 100644
--- a/server/tests/unit/nest/llm-parse/llm-local.service.test.ts
+++ b/server/tests/unit/nest/llm-parse/llm-local.service.test.ts
@@ -68,3 +68,67 @@ describe('LlmLocalService.pull', () => {
expect(JSON.parse(init.body)).toEqual({ model: 'nuextract', stream: true });
});
});
+
+describe('LlmLocalService.modelCapabilities', () => {
+ const showOk = (body: unknown, headers: Record = {}) => ({
+ ok: true,
+ headers: { get: (name: string) => headers[name] ?? null },
+ json: async () => body,
+ });
+
+ it('asks /api/show for the model and answers its capabilities', async () => {
+ const fetchFn = mockFetch(async () => showOk({ capabilities: ['completion', 'vision'] }));
+ await expect(svc().modelCapabilities('http://ollama:11434/v1', 'qwen3.5:4b')).resolves.toEqual(['completion', 'vision']);
+ expect(fetchFn.mock.calls[0][0]).toBe('http://ollama:11434/api/show');
+ expect(JSON.parse(fetchFn.mock.calls[0][1].body)).toEqual({ model: 'qwen3.5:4b' });
+ });
+
+ it('asks the server once per server and model, not once per call', async () => {
+ const fetchFn = mockFetch(async () => showOk({ capabilities: ['completion'] }));
+ const s = svc();
+ await s.modelCapabilities('http://ollama:11434', 'qwen3:8b');
+ await s.modelCapabilities('http://ollama:11434', 'qwen3:8b');
+ expect(fetchFn).toHaveBeenCalledTimes(1);
+ await s.modelCapabilities('http://ollama:11434', 'qwen3.5:4b');
+ expect(fetchFn).toHaveBeenCalledTimes(2);
+ });
+
+ it('answers null for a URL it cannot use, without throwing and without a request', async () => {
+ await expect(svc().modelCapabilities('not a url', 'm')).resolves.toBeNull();
+ await expect(svc().modelCapabilities('ftp://x', 'm')).resolves.toBeNull();
+ expect(safeFetchLlmMock).not.toHaveBeenCalled();
+ });
+
+ it('answers null for an unreachable server, a model it lacks, or a body that is not Ollama\'s', async () => {
+ mockFetch(async () => { throw new Error('ECONNREFUSED'); });
+ await expect(svc().modelCapabilities('http://a:1', 'm')).resolves.toBeNull();
+ mockFetch(async () => ({ ok: false, status: 404, body: null }));
+ await expect(svc().modelCapabilities('http://b:1', 'm')).resolves.toBeNull();
+ mockFetch(async () => showOk({ capabilities: 'vision' }));
+ await expect(svc().modelCapabilities('http://c:1', 'm')).resolves.toBeNull();
+ });
+
+ it('does not read a body that declares itself larger than the cap', async () => {
+ const json = vi.fn(async () => ({ capabilities: ['vision'] }));
+ mockFetch(async () => ({ ...showOk({}), headers: { get: () => String(2 * 1024 * 1024) }, json }));
+ await expect(svc().modelCapabilities('http://d:1', 'm')).resolves.toBeNull();
+ expect(json).not.toHaveBeenCalled();
+ });
+
+ it('asks again a minute after a miss, and keeps an answer for ten', async () => {
+ vi.useFakeTimers();
+ try {
+ const s = svc();
+ mockFetch(async () => { throw new Error('down'); });
+ await s.modelCapabilities('http://e:1', 'm');
+ vi.advanceTimersByTime(61_000);
+ mockFetch(async () => showOk({ capabilities: ['vision'] }));
+ await expect(s.modelCapabilities('http://e:1', 'm')).resolves.toEqual(['vision']);
+ vi.advanceTimersByTime(9 * 60_000);
+ mockFetch(async () => showOk({ capabilities: [] }));
+ await expect(s.modelCapabilities('http://e:1', 'm')).resolves.toEqual(['vision']);
+ } finally {
+ vi.useRealTimers();
+ }
+ });
+});
diff --git a/server/tests/unit/nest/llm-parse/llm-parse.service.test.ts b/server/tests/unit/nest/llm-parse/llm-parse.service.test.ts
index d5ac611e9..cf237c0b7 100644
--- a/server/tests/unit/nest/llm-parse/llm-parse.service.test.ts
+++ b/server/tests/unit/nest/llm-parse/llm-parse.service.test.ts
@@ -10,30 +10,49 @@ const { createLlmClient, extract } = vi.hoisted(() => {
});
vi.mock('../../../../src/nest/llm-parse/llm-client.factory', () => ({ createLlmClient }));
+const { extractEnforced } = vi.hoisted(() => ({ extractEnforced: vi.fn() }));
+
+const { renderPdfPages } = vi.hoisted(() => ({ renderPdfPages: vi.fn() }));
+vi.mock('../../../../src/nest/llm-parse/image-input', async (orig) => {
+ const actual = await orig() as Record;
+ return { ...actual, renderPdfPages };
+});
+vi.mock('../../../../src/nest/llm-parse/router/ollama-format.client', async (orig) => {
+ const actual = await orig() as Record;
+ return { ...actual, extractEnforced };
+});
+
const { extractText } = vi.hoisted(() => ({ extractText: vi.fn(async () => 'Flight AB123') }));
vi.mock('../../../../src/nest/llm-parse/text-extract', async (orig) => {
const actual = await orig() as Record;
return { ...actual, extractText };
});
-const { routeExtraction, detectFlightNumbers } = vi.hoisted(() => ({
+const { routeExtraction, routeImageExtraction, detectFlightNumbers } = vi.hoisted(() => ({
routeExtraction: vi.fn(),
+ routeImageExtraction: vi.fn(),
detectFlightNumbers: vi.fn(() => [] as string[]),
}));
// The router's pure helpers stay real: the currency fallback reads the document's
// total through extractTotalPrice.
vi.mock('../../../../src/nest/llm-parse/router/extraction-router', async (orig) => {
const actual = await orig() as Record;
- return { ...actual, routeExtraction, detectFlightNumbers };
+ return { ...actual, routeExtraction, routeImageExtraction, detectFlightNumbers };
});
import { LlmParseService } from '../../../../src/nest/llm-parse/llm-parse.service';
import type { LlmConfigResolver } from '../../../../src/nest/llm-parse/llm-config.resolver';
import type { RuntimeEnvService } from '../../../../src/nest/app-config/runtime-env.service';
+import type { LlmLocalService } from '../../../../src/nest/llm-parse/llm-local.service';
-const cfg = (over: Record = {}) => ({ provider: 'openai', model: 'm', multimodal: false, ...over });
+const cfg = (over: Record = {}) => ({ provider: 'openai', model: 'm', vision: 'off', ...over });
const llmConfigStub = { resolve: resolveLlmConfig } as unknown as LlmConfigResolver;
-const svc = () => new LlmParseService(llmConfigStub, { isManaged: () => false } as unknown as RuntimeEnvService);
+const modelCapabilities = vi.fn();
+const svc = () => new LlmParseService(
+ llmConfigStub,
+ { isManaged: () => false } as unknown as RuntimeEnvService,
+ { modelCapabilities } as unknown as LlmLocalService,
+);
const file = (name: string, body = 'Flight AB123') => ({ buffer: Buffer.from(body), originalName: name });
beforeEach(() => {
@@ -43,6 +62,9 @@ beforeEach(() => {
extractText.mockResolvedValue('Flight AB123');
detectFlightNumbers.mockReturnValue([]);
routeExtraction.mockResolvedValue({ kiItems: [{ '@type': 'LodgingReservation' }], warnings: [] });
+ routeImageExtraction.mockResolvedValue({ kiItems: [{ '@type': 'TrainReservation' }], warnings: [] });
+ modelCapabilities.mockResolvedValue(null);
+ renderPdfPages.mockResolvedValue([]);
});
describe('LlmParseService', () => {
@@ -284,3 +306,165 @@ describe('LlmParseService: cleaning up an AI answer (#2477)', () => {
expect(res.kiItems).toEqual([plain]);
});
});
+
+describe('LlmParseService: a photo', () => {
+ it('readsImages follows on and off whatever the provider, and never asks a server for them', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ provider: 'local', vision: 'on' }));
+ await expect(svc().readsImages(1)).resolves.toBe(true);
+ resolveLlmConfig.mockReturnValue(cfg({ provider: 'local', vision: 'off' }));
+ await expect(svc().readsImages(1)).resolves.toBe(false);
+ resolveLlmConfig.mockReturnValue(cfg({ provider: 'openai', vision: 'on' }));
+ await expect(svc().readsImages(1)).resolves.toBe(true);
+ expect(modelCapabilities).not.toHaveBeenCalled();
+ });
+
+ it('readsImages on auto asks a local server, and answers no for a cloud provider', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ provider: 'local', vision: 'auto', baseUrl: 'http://ollama:11434/v1', model: 'qwen3.5:4b' }));
+ modelCapabilities.mockResolvedValue(['completion', 'vision']);
+ await expect(svc().readsImages(1)).resolves.toBe(true);
+ expect(modelCapabilities).toHaveBeenCalledWith('http://ollama:11434/v1', 'qwen3.5:4b');
+ modelCapabilities.mockResolvedValue(['completion']);
+ await expect(svc().readsImages(1)).resolves.toBe(false);
+ modelCapabilities.mockResolvedValue(null);
+ await expect(svc().readsImages(1)).resolves.toBe(false);
+
+ modelCapabilities.mockClear();
+ for (const provider of ['openai', 'anthropic']) {
+ resolveLlmConfig.mockReturnValue(cfg({ provider, vision: 'auto' }));
+ await expect(svc().readsImages(1)).resolves.toBe(false);
+ }
+ expect(modelCapabilities).not.toHaveBeenCalled();
+ });
+
+ it('readsImages is false when AI parsing is not configured', async () => {
+ resolveLlmConfig.mockReturnValue(null);
+ await expect(svc().readsImages(1)).resolves.toBe(false);
+ });
+
+ it('refuses a photo the model cannot read, without calling it', async () => {
+ const res = await svc().parse(file('ticket.jpg', 'jpeg'), 1);
+ expect(res.kiItems).toEqual([]);
+ expect(res.warnings[0]).toMatch(/does not read images/);
+ expect(extract).not.toHaveBeenCalled();
+ expect(extractText).not.toHaveBeenCalled();
+ });
+
+ it('sends a photo as an image to a cloud provider, with no text', async () => {
+ for (const provider of ['openai', 'anthropic']) {
+ extract.mockClear();
+ resolveLlmConfig.mockReturnValue(cfg({ provider, vision: 'on' }));
+ await svc().parse(file('ticket.PNG', 'not really a png'), 1);
+ const input = extract.mock.calls[0][0];
+ // Bytes that do not decode go as they came rather than being refused.
+ expect(input.file).toEqual({ mimeType: 'image/png', data: Buffer.from('not really a png') });
+ expect(input.text).toBeUndefined();
+ }
+ expect(extractText).not.toHaveBeenCalled();
+ });
+
+ it('sends a photo through the local router with the image attached', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ provider: 'local', vision: 'on', baseUrl: 'http://ollama:11434/v1' }));
+ const res = await svc().parse(file('ticket.jpeg', 'jpeg bytes'), 1);
+ expect(res.kiItems).toEqual([{ '@type': 'TrainReservation' }]);
+ expect(routeImageExtraction).toHaveBeenCalledWith([Buffer.from('jpeg bytes')], expect.objectContaining({ baseUrl: 'http://ollama:11434/v1' }));
+ expect(routeExtraction).not.toHaveBeenCalled();
+ expect(extract).not.toHaveBeenCalled();
+ });
+
+ it('leaves a PDF on its text path when the model reads images', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ vision: 'on' }));
+ await svc().parse(file('a.pdf', '%PDF'), 1);
+ expect(extract.mock.calls[0][0].file).toBeUndefined();
+ expect(extractText).toHaveBeenCalled();
+ });
+});
+
+describe('LlmParseService.readReceipt', () => {
+ const RECEIPT = { merchant: 'Bäckerei', date: '2026-09-20', total: 7.5, currency: 'EUR', items: [{ name: 'Brezel', price: 2.5 }] };
+
+ it('says why when AI parsing is not set up, the file is not a photo, or the model reads no images', async () => {
+ resolveLlmConfig.mockReturnValue(null);
+ expect((await svc().readReceipt(file('r.jpg'), 1)).warnings[0]).toMatch(/not configured/);
+ resolveLlmConfig.mockReturnValue(cfg({ vision: 'on' }));
+ expect(await svc().readReceipt(file('r.pdf'), 1)).toEqual({ receipt: null, warnings: ['r.pdf: not a photo'] });
+ resolveLlmConfig.mockReturnValue(cfg({ vision: 'off' }));
+ expect((await svc().readReceipt(file('r.jpg'), 1)).warnings[0]).toMatch(/does not read images/);
+ expect(extract).not.toHaveBeenCalled();
+ expect(extractEnforced).not.toHaveBeenCalled();
+ });
+
+ it('reads through Ollama\'s native chat on a local provider, with the photo and the receipt schema', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ provider: 'local', vision: 'on', baseUrl: 'http://ollama:11434/v1', model: 'qwen3.5:4b' }));
+ extractEnforced.mockResolvedValue(RECEIPT);
+ await expect(svc().readReceipt(file('r.jpg', 'jpeg'), 1)).resolves.toEqual({ receipt: RECEIPT, warnings: [] });
+ const call = extractEnforced.mock.calls[0][0];
+ expect(call).toMatchObject({ baseUrl: 'http://ollama:11434/v1', model: 'qwen3.5:4b', images: [Buffer.from('jpeg').toString('base64')] });
+ expect(call.schema.required).toContain('total');
+ expect(extract).not.toHaveBeenCalled();
+ });
+
+ it('asks a cloud client for a receipts list with the photo attached, and keeps the first', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ provider: 'anthropic', vision: 'on' }));
+ extract.mockResolvedValue([RECEIPT, { total: 1 }]);
+ await expect(svc().readReceipt(file('r.png', 'png'), 1)).resolves.toEqual({ receipt: RECEIPT, warnings: [] });
+ const input = extract.mock.calls[0][0];
+ expect(input.rootKey).toBe('receipts');
+ expect(input.file).toEqual({ mimeType: 'image/png', data: Buffer.from('png') });
+ expect(input.text).toBeUndefined();
+ });
+
+ it('answers a warning, not a throw, when the model fails or reads nothing', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ vision: 'on' }));
+ extract.mockRejectedValue(new Error('429'));
+ const failed = await svc().readReceipt(file('r.jpg'), 1);
+ expect(failed.receipt).toBeNull();
+ expect(failed.warnings[0]).toMatch(/AI parsing failed — 429/);
+ extract.mockResolvedValue([]);
+ expect(await svc().readReceipt(file('r.jpg'), 1)).toEqual({ receipt: null, warnings: ['r.jpg: no receipt could be read'] });
+ });
+});
+
+describe('LlmParseService: a scanned PDF', () => {
+ const page = (n: string) => ({ mimeType: 'image/jpeg', data: Buffer.from(n) });
+
+ beforeEach(() => extractText.mockResolvedValue(' '));
+
+ it('sends the pages as images when the text layer is empty and the model reads images', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ vision: 'on' }));
+ renderPdfPages.mockResolvedValue([page('p1'), page('p2')]);
+ await svc().parse(file('scan.pdf', '%PDF'), 1);
+ const input = extract.mock.calls[0][0];
+ expect(input.file).toEqual(page('p1'));
+ expect(input.pageImages).toEqual([page('p2')]);
+ expect(input.text).toBeUndefined();
+ });
+
+ it('hands the local router every page', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ provider: 'local', vision: 'on' }));
+ renderPdfPages.mockResolvedValue([page('p1'), page('p2')]);
+ await svc().parse(file('scan.pdf', '%PDF'), 1);
+ expect(routeImageExtraction).toHaveBeenCalledWith([Buffer.from('p1'), Buffer.from('p2')], expect.anything());
+ expect(routeExtraction).not.toHaveBeenCalled();
+ });
+
+ it('keeps the warning, and draws nothing, when the model reads no images', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ vision: 'off' }));
+ const res = await svc().parse(file('scan.pdf', '%PDF'), 1);
+ expect(res.warnings[0]).toMatch(/no readable text found \(a scanned PDF needs a model that reads images\)/);
+ expect(renderPdfPages).not.toHaveBeenCalled();
+ expect(extract).not.toHaveBeenCalled();
+ });
+
+ it('keeps the warning when no page could be drawn', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ vision: 'on' }));
+ const res = await svc().parse(file('scan.pdf', '%PDF'), 1);
+ expect(res.warnings[0]).toMatch(/no readable text found/);
+ expect(extract).not.toHaveBeenCalled();
+ });
+
+ it('does not draw an empty text file', async () => {
+ resolveLlmConfig.mockReturnValue(cfg({ vision: 'on' }));
+ await svc().parse(file('empty.txt', ''), 1);
+ expect(renderPdfPages).not.toHaveBeenCalled();
+ });
+});
diff --git a/server/tests/unit/nest/llm-parse/ollama-format.client.test.ts b/server/tests/unit/nest/llm-parse/ollama-format.client.test.ts
index 3d56541e7..60b5eaa6d 100644
--- a/server/tests/unit/nest/llm-parse/ollama-format.client.test.ts
+++ b/server/tests/unit/nest/llm-parse/ollama-format.client.test.ts
@@ -52,6 +52,15 @@ describe('extractEnforced', () => {
expect((init as RequestInit).headers).not.toHaveProperty('authorization');
});
+ it('attaches images to the user turn only when given', async () => {
+ const fetchFn = mockFetch(() => jsonResponse({ message: { content: '{}' } }));
+ await extractEnforced({ ...INPUT, images: ['aW1n'] });
+ await extractEnforced(INPUT);
+ const userTurn = (i: number) => JSON.parse((fetchFn.mock.calls[i][1] as RequestInit).body as string).messages[1];
+ expect(userTurn(0).images).toEqual(['aW1n']);
+ expect(userTurn(1)).not.toHaveProperty('images');
+ });
+
it('sends a bearer header only when an apiKey is given', async () => {
const fetchFn = mockFetch(() => jsonResponse({ message: { content: '{}' } }));
await extractEnforced({ ...INPUT, apiKey: 'sk-123', numPredict: 900, numCtx: 16000 });
diff --git a/server/tests/unit/nest/llm-parse/receipt-read.test.ts b/server/tests/unit/nest/llm-parse/receipt-read.test.ts
new file mode 100644
index 000000000..6485dded1
--- /dev/null
+++ b/server/tests/unit/nest/llm-parse/receipt-read.test.ts
@@ -0,0 +1,52 @@
+import { describe, it, expect } from 'vitest';
+import { buildReceiptPrompt, RECEIPT_LIST_JSON_SCHEMA, RECEIPT_ROOT_KEY, toReceiptRead } from '../../../../src/nest/llm-parse/receipt-read';
+
+describe('toReceiptRead', () => {
+ it('keeps a clean answer as it came', () => {
+ expect(toReceiptRead({
+ merchant: 'Café de Flore', date: '2026-09-20', total: 23.4, currency: 'EUR',
+ items: [{ name: '2 x Espresso', price: 8 }, { name: 'Croque', price: 15.4 }],
+ })).toEqual({
+ merchant: 'Café de Flore', date: '2026-09-20', total: 23.4, currency: 'EUR',
+ items: [{ name: '2 x Espresso', price: 8 }, { name: 'Croque', price: 15.4 }],
+ });
+ });
+
+ it('reads amounts written as text, in either decimal convention', () => {
+ expect(toReceiptRead({ total: '12,50 €' })?.total).toBe(12.5);
+ expect(toReceiptRead({ total: '1.234,50' })?.total).toBe(1234.5);
+ expect(toReceiptRead({ total: '1,234.50' })?.total).toBe(1234.5);
+ expect(toReceiptRead({ total: 7, items: [{ name: 'Tea', price: '3,20' }] })?.items).toEqual([{ name: 'Tea', price: 3.2 }]);
+ });
+
+ it('turns a currency symbol or name into its code, and drops one nobody can convert', () => {
+ expect(toReceiptRead({ total: 5, currency: '€' })?.currency).toBe('EUR');
+ expect(toReceiptRead({ total: 5, currency: 'euros' })?.currency).toBe('EUR');
+ expect(toReceiptRead({ total: 5, currency: 'Lek' })?.currency).toBeNull();
+ });
+
+ it('keeps only a real calendar date', () => {
+ expect(toReceiptRead({ total: 5, date: '2026-02-28' })?.date).toBe('2026-02-28');
+ for (const date of ['2026-02-30', '28/02/2026', '2026-2-8', '']) expect(toReceiptRead({ total: 5, date })?.date).toBeNull();
+ });
+
+ it('leaves out a total that is not a positive amount, and lines without a name or price', () => {
+ const read = toReceiptRead({ merchant: 'Shop', total: 0, items: [{ name: '', price: 2 }, { name: 'Bread' }, 'x', { name: 'Milk', price: 1.1 }] });
+ expect(read).toEqual({ merchant: 'Shop', date: null, total: null, currency: null, items: [{ name: 'Milk', price: 1.1 }] });
+ });
+
+ it('is null when nothing usable came back', () => {
+ for (const raw of [null, undefined, 'text', [], {}, { total: 'n/a', merchant: ' ' }]) expect(toReceiptRead(raw)).toBeNull();
+ });
+});
+
+describe('the receipt prompt and schema', () => {
+ it('pins today so an ambiguous date is read towards it', () => {
+ expect(buildReceiptPrompt(new Date('2026-09-24T10:00:00Z'))).toContain('Today is 2026-09-24.');
+ });
+
+ it('wraps one receipt under the key the cloud clients read', () => {
+ expect(RECEIPT_LIST_JSON_SCHEMA.required).toEqual([RECEIPT_ROOT_KEY]);
+ expect(RECEIPT_LIST_JSON_SCHEMA.properties[RECEIPT_ROOT_KEY].items.required).toEqual(['merchant', 'date', 'total', 'currency', 'items']);
+ });
+});
diff --git a/server/tests/unit/nest/receipt-scan.controller.test.ts b/server/tests/unit/nest/receipt-scan.controller.test.ts
new file mode 100644
index 000000000..7335db46b
--- /dev/null
+++ b/server/tests/unit/nest/receipt-scan.controller.test.ts
@@ -0,0 +1,55 @@
+import { describe, it, expect, vi } from 'vitest';
+import { HttpException } from '@nestjs/common';
+import { ReceiptScanController } from '../../../src/nest/receipt-scan/receipt-scan.controller';
+import type { ImportJobsService } from '../../../src/nest/booking-import/import-jobs.service';
+import type { LlmParseService } from '../../../src/nest/llm-parse/llm-parse.service';
+import type { DatabaseService } from '../../../src/nest/database/database.service';
+import type { PermissionsService } from '../../../src/nest/permissions/permissions.service';
+import type { AddonsService } from '../../../src/nest/addons/addons.service';
+import type { User } from '../../../src/types';
+
+const user = { id: 3, role: 'user' } as User;
+const photo = (name = 'r.jpg') => ({ originalname: name, buffer: Buffer.from('x') } as Express.Multer.File);
+
+function make(over: { readsImages?: boolean; trip?: unknown; addon?: boolean; allowed?: boolean } = {}) {
+ const startReceipt = vi.fn(() => 'job-1');
+ const checkPermission = vi.fn(() => over.allowed ?? true);
+ const c = new ReceiptScanController(
+ { startReceipt } as unknown as ImportJobsService,
+ { readsImages: vi.fn(async () => over.readsImages ?? true) } as unknown as LlmParseService,
+ { canAccessTrip: vi.fn(() => ('trip' in over ? over.trip : { user_id: 5 })) } as unknown as DatabaseService,
+ { checkPermission } as unknown as PermissionsService,
+ { isAddonEnabled: vi.fn(() => over.addon ?? true) } as unknown as AddonsService,
+ );
+ return { c, startReceipt, checkPermission };
+}
+
+async function failure(fn: () => Promise): Promise<[number, unknown]> {
+ try { await fn(); } catch (e) { expect(e).toBeInstanceOf(HttpException); return [(e as HttpException).getStatus(), (e as HttpException).getResponse()]; }
+ throw new Error('expected throw');
+}
+
+describe('ReceiptScanController', () => {
+ it('queues a photo and answers the job id', async () => {
+ const { c, startReceipt, checkPermission } = make();
+ await expect(c.scan(user, '9', photo())).resolves.toEqual({ jobId: 'job-1' });
+ expect(startReceipt).toHaveBeenCalledWith('9', expect.objectContaining({ originalname: 'r.jpg' }), 3);
+ expect(checkPermission).toHaveBeenCalledWith('budget_edit', 'user', 5, 3, true);
+ });
+
+ it('answers what the guards would: 404 for a trip out of reach or Costs off, 403 without budget_edit', async () => {
+ expect(await failure(() => make({ trip: undefined }).c.scan(user, '9', photo()))).toEqual([404, { error: 'Trip not found' }]);
+ expect(await failure(() => make({ addon: false }).c.scan(user, '9', photo()))).toEqual([404, { error: 'Costs addon is not enabled' }]);
+ const { c, startReceipt } = make({ allowed: false });
+ expect(await failure(() => c.scan(user, '9', photo()))).toEqual([403, { error: 'No permission' }]);
+ expect(startReceipt).not.toHaveBeenCalled();
+ });
+
+ it('refuses with 400 a missing file, a file that is not a photo, or a model that reads no images', async () => {
+ expect((await failure(() => make().c.scan(user, '9', undefined)))[0]).toBe(400);
+ expect((await failure(() => make().c.scan(user, '9', photo('r.pdf'))))[0]).toBe(400);
+ const { c, startReceipt } = make({ readsImages: false });
+ expect(await failure(() => c.scan(user, '9', photo()))).toEqual([400, { error: 'The configured AI model does not read photos' }]);
+ expect(startReceipt).not.toHaveBeenCalled();
+ });
+});
diff --git a/shared/src/admin/admin.schema.spec.ts b/shared/src/admin/admin.schema.spec.ts
index b3b6d8a81..677f5d788 100644
--- a/shared/src/admin/admin.schema.spec.ts
+++ b/shared/src/admin/admin.schema.spec.ts
@@ -1,4 +1,6 @@
import {
+ asLlmVision,
+ LLM_VISION_MODES,
adminUserCreateRequestSchema,
adminUserUpdateRequestSchema,
adminPermissionsRequestSchema,
@@ -106,6 +108,13 @@ describe('adminAddonUpdateRequestSchema', () => {
});
});
+describe('asLlmVision', () => {
+ it('keeps the three modes and reads anything else as auto', () => {
+ for (const mode of LLM_VISION_MODES) expect(asLlmVision(mode)).toBe(mode);
+ for (const other of [undefined, null, '', 'ON', true, false, 1]) expect(asLlmVision(other)).toBe('auto');
+ });
+});
+
describe('adminCollabFeaturesRequestSchema', () => {
it('accepts a single computed key', () => {
expect(adminCollabFeaturesRequestSchema.safeParse({ chat: false }).success).toBe(true);
diff --git a/shared/src/admin/admin.schema.ts b/shared/src/admin/admin.schema.ts
index fb6c2d9e1..26f27670b 100644
--- a/shared/src/admin/admin.schema.ts
+++ b/shared/src/admin/admin.schema.ts
@@ -80,6 +80,30 @@ export const adminAddonUpdateRequestSchema = z.object({
});
export type AdminAddonUpdateRequest = z.infer;
+/**
+ * Whether the AI Parsing model is handed a photo, stored as `vision` on that
+ * addon's config. `auto` asks a local Ollama server what the model can do and
+ * means no for a cloud provider; `on` and `off` are the admin's word.
+ */
+export const LLM_VISION_MODES = ['auto', 'on', 'off'] as const;
+export type LlmVision = (typeof LLM_VISION_MODES)[number];
+
+/**
+ * The photos handed to a model that reads images. HEIC is not here: no provider
+ * TREK talks to reads it, and a browser picking from the camera roll hands over
+ * a JPEG anyway.
+ */
+export const LLM_PHOTO_EXTENSIONS = ['.jpg', '.jpeg', '.png', '.webp'] as const;
+
+/** GET /api/llm/capabilities: what the caller's AI model can be handed. */
+export const llmCapabilitiesResponseSchema = z.object({ images: z.boolean() });
+export type LlmCapabilitiesResponse = z.infer;
+
+/** A stored or submitted value as a mode; anything else, a missing value included, is `auto`. */
+export function asLlmVision(value: unknown): LlmVision {
+ return (LLM_VISION_MODES as readonly unknown[]).includes(value) ? (value as LlmVision) : 'auto';
+}
+
// Fully partial — the client PUTs one computed key at a time.
export const adminCollabFeaturesRequestSchema = z.object({
chat: z.boolean().optional(),
diff --git a/shared/src/budget/receipt-scan.schema.ts b/shared/src/budget/receipt-scan.schema.ts
new file mode 100644
index 000000000..00ab391c7
--- /dev/null
+++ b/shared/src/budget/receipt-scan.schema.ts
@@ -0,0 +1,46 @@
+import { z } from 'zod';
+
+/**
+ * A photographed receipt, as the AI model read it for the Costs tab.
+ *
+ * It only pre-fills the expense editor: nothing is saved until the person saves
+ * the expense there, so every field may be missing and each is checked by eye.
+ * There is no category: which kind of spending it was is the person's call.
+ */
+export const receiptLineSchema = z.object({
+ name: z.string().min(1),
+ /** What the line cost in total, quantity included, in the receipt's currency. */
+ price: z.number(),
+});
+export type ReceiptLine = z.infer;
+
+export const receiptReadSchema = z.object({
+ /** The business, which becomes the expense name. */
+ merchant: z.string().min(1).nullable(),
+ /** YYYY-MM-DD. */
+ date: z
+ .string()
+ .regex(/^\d{4}-\d{2}-\d{2}$/)
+ .nullable(),
+ /** The grand total paid. */
+ total: z.number().positive().nullable(),
+ /** ISO 4217. */
+ currency: z
+ .string()
+ .regex(/^[A-Z]{3}$/)
+ .nullable(),
+ /** The itemized lines, which seed the Ticket split. */
+ items: z.array(receiptLineSchema),
+});
+export type ReceiptRead = z.infer;
+
+/** What a finished receipt scan job answers: the read, or null with the reason in `warnings`. */
+export const receiptScanResultSchema = z.object({
+ receipt: receiptReadSchema.nullable(),
+ warnings: z.array(z.string()),
+});
+export type ReceiptScanResult = z.infer;
+
+/** POST /api/trips/:tripId/budget/receipt-scan: the background job reading the photo. */
+export const receiptScanStartResponseSchema = z.object({ jobId: z.string().min(1) });
+export type ReceiptScanStartResponse = z.infer;
diff --git a/shared/src/i18n/ar/admin.ts b/shared/src/i18n/ar/admin.ts
index 47fdf2043..ed808c599 100644
--- a/shared/src/i18n/ar/admin.ts
+++ b/shared/src/i18n/ar/admin.ts
@@ -626,6 +626,11 @@ const admin: TranslationStrings = {
'قراءة الزيارات والمسارات المسجَّلة من خادم Dawarich يربطه كل قارئ بنفسه',
'admin.addons.catalog.llm_parsing.name': 'التحليل بالذكاء الاصطناعي',
'admin.addons.catalog.llm_parsing.description': 'يقرأ الحجوزات التي يعجز المحلل المدمج عن فهمها، بنموذج ذكاء اصطناعي تختاره',
+ 'admin.addons.llm.vision.auto': 'تلقائي',
+ 'admin.addons.llm.vision.on': 'نعم',
+ 'admin.addons.llm.vision.off': 'لا',
+ 'admin.addons.llm.vision.hintLocal': 'يسأل الخيار «تلقائي» خادم Ollama عمّا إذا كان هذا النموذج يقرأ الصور.',
+ 'admin.addons.llm.vision.hintCloud': 'يعني «تلقائي» «لا» للنموذج السحابي. اختر «نعم» إذا كان هذا النموذج يقرأ الصور.',
'admin.addons.enabled': 'مفعّل',
'admin.addons.disabled': 'معطّل',
'admin.addons.type.trip': 'رحلة',
diff --git a/shared/src/i18n/ar/budget.ts b/shared/src/i18n/ar/budget.ts
index 3174e063c..5139e9709 100644
--- a/shared/src/i18n/ar/budget.ts
+++ b/shared/src/i18n/ar/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'الإيصالات',
'costs.receiptsTitle': 'الإيصالات والفواتير',
'costs.attachReceipt': 'إرفاق إيصال / فاتورة',
+ 'costs.scan.button': 'مسح إيصال',
+ 'costs.scan.reading': 'جارٍ قراءة الإيصال…',
+ 'costs.scan.review': 'مراجعة المصروف',
+ 'costs.scan.nothingRead': 'تعذّرت قراءة أي إيصال من هذه الصورة.',
+ 'costs.scan.failed': 'تعذّر بدء قراءة الإيصال.',
+ 'costs.scan.title': 'مسح إيصال',
+ 'costs.scan.accepted': 'صورة لإيصال واحد (JPG أو PNG أو WEBP، حتى 10 ميغابايت). يقرأها نموذج الذكاء الاصطناعي في الخلفية، ثم تراجع المصروف قبل حفظه.',
+ 'costs.scan.dropHere': 'أفلت صورة الإيصال هنا، أو انقر لاختيار صورة أو التقاطها',
+ 'costs.scan.dropActive': 'أفلت الصورة لمسحها',
+ 'costs.scan.start': 'مسح',
+ 'costs.scan.unsupported': 'يمكن مسح الصور فقط: JPG أو PNG أو WEBP.',
'costs.noReceipts': 'لا توجد إيصالات مرفقة',
'costs.deleteReceipt': 'حذف الإيصال',
'costs.viewReceipt': 'عرض الإيصال',
diff --git a/shared/src/i18n/ar/reservations.ts b/shared/src/i18n/ar/reservations.ts
index ce24ec4ab..714da1379 100644
--- a/shared/src/i18n/ar/reservations.ts
+++ b/shared/src/i18n/ar/reservations.ts
@@ -129,6 +129,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'أسقط الملفات للاستيراد',
'reservations.import.acceptedFormats':
'المقبول: EML، PDF، PKPass، HTML، TXT (بحد أقصى 10 ميغابايت لكل ملف، حتى 5 ملفات)',
+ 'reservations.import.acceptedPhotos': 'يقرأ نموذج الذكاء الاصطناعي الصور (JPG وPNG وWEBP).',
'reservations.import.parsing': 'جارٍ معالجة الملفات…',
'reservations.import.previewHeading': 'تم العثور على {count} حجز/حجوزات',
'reservations.import.previewEmpty': 'تعذّر استخراج أي حجوزات من الملفات المُحمَّلة.',
diff --git a/shared/src/i18n/ar/settings.ts b/shared/src/i18n/ar/settings.ts
index 213936d27..a97c52adb 100644
--- a/shared/src/i18n/ar/settings.ts
+++ b/shared/src/i18n/ar/settings.ts
@@ -383,8 +383,8 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'المكان الذي يعمل فيه النموذج — خادم Ollama محلي أو نقطة نهاية متوافقة مع OpenAI.',
'settings.aiParsing.apiKey': 'مفتاح API',
'settings.aiParsing.apiKeyHint': 'يُخزَّن مشفّرًا. اتركه فارغًا للإبقاء على المفتاح الحالي.',
- 'settings.aiParsing.multimodal': 'إرسال المستندات كصور',
- 'settings.aiParsing.multimodalHint': 'للنماذج القادرة على الرؤية — يرسل ملف PDF الأصلي بدلًا من النص المستخرج.',
+ 'settings.aiParsing.multimodal': 'النموذج يقرأ الصور',
+ 'settings.aiParsing.multimodalHint': 'فعّله لنموذج يقرأ الصور، ليمكن استيراد صورة أو مسحها.',
'settings.aiParsing.toast.saved': 'تم حفظ إعدادات الذكاء الاصطناعي',
'settings.aiParsing.toast.saveError': 'تعذّر حفظ إعدادات الذكاء الاصطناعي',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/az/admin.ts b/shared/src/i18n/az/admin.ts
index d022a4561..b96b142dc 100644
--- a/shared/src/i18n/az/admin.ts
+++ b/shared/src/i18n/az/admin.ts
@@ -733,6 +733,11 @@ const admin: TranslationStrings = {
'admin.addons.catalog.llm_parsing.name': 'Süni intellektlə təhlil',
'admin.addons.catalog.llm_parsing.description':
'Daxili təhlilçinin oxuya bilmədiyi rezervasiyaları seçdiyiniz süni intellekt modeli ilə oxuyur',
+ 'admin.addons.llm.vision.auto': 'Avtomatik',
+ 'admin.addons.llm.vision.on': 'Bəli',
+ 'admin.addons.llm.vision.off': 'Xeyr',
+ 'admin.addons.llm.vision.hintLocal': 'Avtomatik rejim bu modelin şəkilləri oxuyub-oxumadığını Ollama serverindən soruşur.',
+ 'admin.addons.llm.vision.hintCloud': 'Bulud modeli üçün Avtomatik "xeyr" deməkdir. Bu model şəkilləri oxuyursa, Bəli seç.',
'admin.addons.enabled': 'Aktivdir',
'admin.addons.disabled': 'Deaktivdir',
'admin.addons.type.trip': 'Səyahət',
diff --git a/shared/src/i18n/az/budget.ts b/shared/src/i18n/az/budget.ts
index a2735133c..401254cce 100644
--- a/shared/src/i18n/az/budget.ts
+++ b/shared/src/i18n/az/budget.ts
@@ -186,6 +186,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Qəbzlər',
'costs.receiptsTitle': 'Qəbzlər və fakturalar',
'costs.attachReceipt': 'Qəbz / faktura əlavə et',
+ 'costs.scan.button': 'Qəbzi skan et',
+ 'costs.scan.reading': 'Qəbz oxunur…',
+ 'costs.scan.review': 'Xərci yoxla',
+ 'costs.scan.nothingRead': 'Bu fotodan qəbz oxunmadı.',
+ 'costs.scan.failed': 'Qəbzin oxunmasına başlamaq mümkün olmadı.',
+ 'costs.scan.title': 'Qəbzi skan et',
+ 'costs.scan.accepted': 'Bir qəbzin fotosu (JPG, PNG və ya WEBP, 10 MB-a qədər). Süni intellekt modeli onu arxa planda oxuyur, sonra xərci saxlamazdan əvvəl yoxlayırsan.',
+ 'costs.scan.dropHere': 'Qəbzin fotosunu bura burax və ya seçmək ya çəkmək üçün kliklə',
+ 'costs.scan.dropActive': 'Skan etmək üçün fotonu burax',
+ 'costs.scan.start': 'Skan et',
+ 'costs.scan.unsupported': 'Yalnız foto skan edilə bilər: JPG, PNG və ya WEBP.',
'costs.noReceipts': 'Əlavə edilmiş qəbz yoxdur',
'costs.deleteReceipt': 'Qəbzi sil',
'costs.viewReceipt': 'Qəbzə bax',
diff --git a/shared/src/i18n/az/reservations.ts b/shared/src/i18n/az/reservations.ts
index dcfa39e07..6d356a854 100644
--- a/shared/src/i18n/az/reservations.ts
+++ b/shared/src/i18n/az/reservations.ts
@@ -179,6 +179,7 @@ const reservations: TranslationStrings = {
'İdxal etmək üçün faylları buraya buraxın',
'reservations.import.acceptedFormats':
'Qəbul edilən formatlar: EML, PDF, PKPass, HTML, TXT (hər biri maksimum 10 MB, 5 fayladək)',
+ 'reservations.import.acceptedPhotos': 'Fotoları (JPG, PNG, WEBP) süni intellekt modeli oxuyur.',
'reservations.import.parsing': 'Fayllar təhlil edilir…',
'reservations.import.previewHeading':
'{count} rezervasiya tapıldı',
diff --git a/shared/src/i18n/az/settings.ts b/shared/src/i18n/az/settings.ts
index 7cadf3bdd..195ffa261 100644
--- a/shared/src/i18n/az/settings.ts
+++ b/shared/src/i18n/az/settings.ts
@@ -497,9 +497,9 @@ const settings: TranslationStrings = {
'settings.aiParsing.apiKey': 'API açarı',
'settings.aiParsing.apiKeyHint':
'Şifrələnmiş şəkildə saxlanılır. Cari açarı saxlamaq üçün boş buraxın.',
- 'settings.aiParsing.multimodal': 'Sənədləri şəkil kimi göndər',
+ 'settings.aiParsing.multimodal': 'Model şəkilləri oxuyur',
'settings.aiParsing.multimodalHint':
- 'Görmə qabiliyyətli modellər üçün — çıxarılmış mətn əvəzinə orijinal PDF-i göndərir.',
+ 'Şəkilləri oxuyan model üçün aç: onda foto idxal oluna və ya skan edilə bilər.',
'settings.aiParsing.toast.saved': 'AI tənzimləmələri yadda saxlanıldı',
'settings.aiParsing.toast.saveError':
'AI tənzimləmələrini yadda saxlamaq mümkün olmadı',
diff --git a/shared/src/i18n/br/admin.ts b/shared/src/i18n/br/admin.ts
index bdd6b6f0e..da723c3ae 100644
--- a/shared/src/i18n/br/admin.ts
+++ b/shared/src/i18n/br/admin.ts
@@ -618,6 +618,11 @@ const admin: TranslationStrings = {
'Leia visitas e rotas gravadas de uma instância do Dawarich que cada leitor conecta por conta própria',
'admin.addons.catalog.llm_parsing.name': 'Análise por IA',
'admin.addons.catalog.llm_parsing.description': 'Lê reservas que o analisador integrado não entende, usando um modelo de IA à sua escolha',
+ 'admin.addons.llm.vision.auto': 'Automático',
+ 'admin.addons.llm.vision.on': 'Sim',
+ 'admin.addons.llm.vision.off': 'Não',
+ 'admin.addons.llm.vision.hintLocal': 'Automático pergunta ao servidor Ollama se este modelo lê imagens.',
+ 'admin.addons.llm.vision.hintCloud': 'Automático significa não para um modelo na nuvem. Escolha Sim se este modelo lê imagens.',
'admin.addons.enabled': 'Ativado',
'admin.addons.disabled': 'Desativado',
'admin.addons.type.trip': 'Viagem',
diff --git a/shared/src/i18n/br/budget.ts b/shared/src/i18n/br/budget.ts
index da0159ac4..b685ee821 100644
--- a/shared/src/i18n/br/budget.ts
+++ b/shared/src/i18n/br/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Comprovantes',
'costs.receiptsTitle': 'Comprovantes e notas',
'costs.attachReceipt': 'Anexar comprovante / nota',
+ 'costs.scan.button': 'Escanear comprovante',
+ 'costs.scan.reading': 'Lendo o comprovante…',
+ 'costs.scan.review': 'Revisar a despesa',
+ 'costs.scan.nothingRead': 'Não foi possível ler um comprovante nesta foto.',
+ 'costs.scan.failed': 'Não foi possível começar a ler o comprovante.',
+ 'costs.scan.title': 'Escanear comprovante',
+ 'costs.scan.accepted': 'A foto de um único comprovante (JPG, PNG ou WEBP, até 10 MB). O modelo de IA lê em segundo plano e depois você confere a despesa antes de salvar.',
+ 'costs.scan.dropHere': 'Solte aqui a foto do comprovante, ou clique para escolher ou tirar uma',
+ 'costs.scan.dropActive': 'Solte a foto para escaneá-la',
+ 'costs.scan.start': 'Escanear',
+ 'costs.scan.unsupported': 'Só é possível escanear uma foto: JPG, PNG ou WEBP.',
'costs.noReceipts': 'Nenhum comprovante anexado',
'costs.deleteReceipt': 'Remover comprovante',
'costs.viewReceipt': 'Ver comprovante',
diff --git a/shared/src/i18n/br/reservations.ts b/shared/src/i18n/br/reservations.ts
index ff1856544..10ff7f5fa 100644
--- a/shared/src/i18n/br/reservations.ts
+++ b/shared/src/i18n/br/reservations.ts
@@ -128,6 +128,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Solte os arquivos de confirmação de reserva aqui ou clique para selecionar',
'reservations.import.dropActive': 'Solte os arquivos para importar',
'reservations.import.acceptedFormats': 'Aceitos: EML, PDF, PKPass, HTML, TXT (máx. 10 MB cada, até 5 arquivos)',
+ 'reservations.import.acceptedPhotos': 'As fotos (JPG, PNG, WEBP) são lidas pelo modelo de IA.',
'reservations.import.parsing': 'Analisando arquivos…',
'reservations.import.previewHeading': '{count} reserva(s) encontrada(s)',
'reservations.import.previewEmpty': 'Nenhuma reserva pôde ser extraída dos arquivos enviados.',
diff --git a/shared/src/i18n/br/settings.ts b/shared/src/i18n/br/settings.ts
index a81eabad9..054c19b97 100644
--- a/shared/src/i18n/br/settings.ts
+++ b/shared/src/i18n/br/settings.ts
@@ -398,9 +398,9 @@ const settings: TranslationStrings = {
'Onde o modelo é executado — um servidor Ollama local ou um endpoint compatível com OpenAI.',
'settings.aiParsing.apiKey': 'Chave de API',
'settings.aiParsing.apiKeyHint': 'Armazenada de forma criptografada. Deixe em branco para manter a chave atual.',
- 'settings.aiParsing.multimodal': 'Enviar documentos como imagens',
+ 'settings.aiParsing.multimodal': 'O modelo lê imagens',
'settings.aiParsing.multimodalHint':
- 'Para modelos com capacidade de visão — envia o PDF original em vez do texto extraído.',
+ 'Ative para um modelo que lê imagens, assim uma foto pode ser importada ou escaneada.',
'settings.aiParsing.toast.saved': 'Configurações de IA salvas',
'settings.aiParsing.toast.saveError': 'Não foi possível salvar as configurações de IA',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/ca/admin.ts b/shared/src/i18n/ca/admin.ts
index e82b372e3..f8553dfb1 100644
--- a/shared/src/i18n/ca/admin.ts
+++ b/shared/src/i18n/ca/admin.ts
@@ -233,6 +233,11 @@ const admin: TranslationStrings = {
'Llegeix estades i recorreguts enregistrats d’una instància de Dawarich que cada lector connecta pel seu compte',
'admin.addons.catalog.llm_parsing.name': 'Anàlisi amb IA',
'admin.addons.catalog.llm_parsing.description': 'Llegeix reserves que l’analitzador integrat no entén, amb un model d’IA que triïs',
+ 'admin.addons.llm.vision.auto': 'Automàtic',
+ 'admin.addons.llm.vision.on': 'Sí',
+ 'admin.addons.llm.vision.off': 'No',
+ 'admin.addons.llm.vision.hintLocal': 'Automàtic pregunta al servidor Ollama si aquest model llegeix imatges.',
+ 'admin.addons.llm.vision.hintCloud': 'Automàtic vol dir no per a un model al núvol. Tria Sí si aquest model llegeix imatges.',
'admin.addons.enabled': 'Actiu',
'admin.addons.disabled': 'Desactivat',
'admin.addons.type.trip': 'Viatge',
diff --git a/shared/src/i18n/ca/budget.ts b/shared/src/i18n/ca/budget.ts
index e4ca7ee04..16cd6e5dc 100644
--- a/shared/src/i18n/ca/budget.ts
+++ b/shared/src/i18n/ca/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Rebuts',
'costs.receiptsTitle': 'Rebuts i factures',
'costs.attachReceipt': 'Adjuntar rebut / factura',
+ 'costs.scan.button': 'Escanejar rebut',
+ 'costs.scan.reading': 'Llegint el rebut…',
+ 'costs.scan.review': 'Revisar la despesa',
+ 'costs.scan.nothingRead': "No s'ha pogut llegir cap rebut d'aquesta foto.",
+ 'costs.scan.failed': "No s'ha pogut començar a llegir el rebut.",
+ 'costs.scan.title': 'Escanejar rebut',
+ 'costs.scan.accepted': "Una foto d'un sol rebut (JPG, PNG o WEBP, fins a 10 MB). El model d'IA la llegeix en segon pla i després revises la despesa abans de desar-la.",
+ 'costs.scan.dropHere': 'Deixa aquí la foto del rebut, o fes clic per triar-ne o fer-ne una',
+ 'costs.scan.dropActive': 'Deixa anar la foto per escanejar-la',
+ 'costs.scan.start': 'Escanejar',
+ 'costs.scan.unsupported': 'Només es pot escanejar una foto: JPG, PNG o WEBP.',
'costs.noReceipts': 'Cap rebut adjunt',
'costs.deleteReceipt': 'Eliminar rebut',
'costs.viewReceipt': 'Veure rebut',
diff --git a/shared/src/i18n/ca/reservations.ts b/shared/src/i18n/ca/reservations.ts
index c7e83153e..7cb7f7042 100644
--- a/shared/src/i18n/ca/reservations.ts
+++ b/shared/src/i18n/ca/reservations.ts
@@ -129,6 +129,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Deixa anar els fitxers per importar',
'reservations.import.acceptedFormats':
'Acceptats: EML, PDF, PKPass, HTML, TXT (màx. 10 MB per fitxer, fins a 5 fitxers)',
+ 'reservations.import.acceptedPhotos': "El model d'IA llegeix les fotos (JPG, PNG, WEBP).",
'reservations.import.parsing': 'Analitzant els fitxers…',
'reservations.import.previewHeading': '{count} reserva(es) trobada(es)',
'reservations.import.previewEmpty': "No s'han pogut extreure reserves dels fitxers pujats.",
diff --git a/shared/src/i18n/ca/settings.ts b/shared/src/i18n/ca/settings.ts
index b1faebc43..9e05533fe 100644
--- a/shared/src/i18n/ca/settings.ts
+++ b/shared/src/i18n/ca/settings.ts
@@ -379,9 +379,9 @@ const settings: TranslationStrings = {
"On s'executa el model — un servidor local d'Ollama o un punt final compatible amb OpenAI.",
'settings.aiParsing.apiKey': "Clau de l'API",
'settings.aiParsing.apiKeyHint': "S'emmagatzema xifrada. Deixa-ho en blanc per mantenir la clau actual.",
- 'settings.aiParsing.multimodal': 'Enviar documents com a imatges',
+ 'settings.aiParsing.multimodal': 'El model llegeix imatges',
'settings.aiParsing.multimodalHint':
- 'Per a models amb capacitats de visió — envia el PDF original en comptes del text extret.',
+ 'Activa-ho per a un model que llegeix imatges, així es pot importar o escanejar una foto.',
'settings.aiParsing.toast.saved': "Ajustos d'IA desats",
'settings.aiParsing.toast.saveError': "No s'han pogut desar els ajustos d'IA",
'settings.tabs.appearance': 'Aparença',
diff --git a/shared/src/i18n/cs/admin.ts b/shared/src/i18n/cs/admin.ts
index b8f4fc217..781bf89c9 100644
--- a/shared/src/i18n/cs/admin.ts
+++ b/shared/src/i18n/cs/admin.ts
@@ -608,6 +608,11 @@ const admin: TranslationStrings = {
'Načtěte návštěvy a zaznamenané trasy z instance Dawarich, kterou si každý čtenář připojí sám',
'admin.addons.catalog.llm_parsing.name': 'Analýza pomocí AI',
'admin.addons.catalog.llm_parsing.description': 'Přečte rezervace, které vestavěný analyzátor nezvládne, pomocí zvoleného modelu AI',
+ 'admin.addons.llm.vision.auto': 'Automaticky',
+ 'admin.addons.llm.vision.on': 'Ano',
+ 'admin.addons.llm.vision.off': 'Ne',
+ 'admin.addons.llm.vision.hintLocal': 'Automaticky se zeptá serveru Ollama, zda tento model čte obrázky.',
+ 'admin.addons.llm.vision.hintCloud': 'Automaticky znamená u cloudového modelu ne. Zvolte Ano, pokud tento model čte obrázky.',
'admin.addons.catalog.roadtrip.name': 'Cesta autem',
'admin.addons.catalog.roadtrip.description': 'Plánujte jízdy se zastávkami – doby jízdy a časy příjezdu se přepočítají samy',
'admin.addons.enabled': 'Povoleno',
diff --git a/shared/src/i18n/cs/budget.ts b/shared/src/i18n/cs/budget.ts
index da498a246..b928b2382 100644
--- a/shared/src/i18n/cs/budget.ts
+++ b/shared/src/i18n/cs/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Účtenky',
'costs.receiptsTitle': 'Účtenky a faktury',
'costs.attachReceipt': 'Připojit účtenku / fakturu',
+ 'costs.scan.button': 'Naskenovat účtenku',
+ 'costs.scan.reading': 'Čtení účtenky…',
+ 'costs.scan.review': 'Zkontrolovat výdaj',
+ 'costs.scan.nothingRead': 'Z této fotografie se nepodařilo přečíst žádnou účtenku.',
+ 'costs.scan.failed': 'Čtení účtenky se nepodařilo spustit.',
+ 'costs.scan.title': 'Naskenovat účtenku',
+ 'costs.scan.accepted': 'Fotografie jedné účtenky (JPG, PNG nebo WEBP, do 10 MB). Model AI ji přečte na pozadí a vy pak výdaj před uložením zkontrolujete.',
+ 'costs.scan.dropHere': 'Přetáhněte sem fotografii účtenky, nebo klikněte a vyberte či pořiďte novou',
+ 'costs.scan.dropActive': 'Pusťte fotografii pro naskenování',
+ 'costs.scan.start': 'Skenovat',
+ 'costs.scan.unsupported': 'Skenovat lze jen fotografii: JPG, PNG nebo WEBP.',
'costs.noReceipts': 'Žádné připojené účtenky',
'costs.deleteReceipt': 'Odebrat účtenku',
'costs.viewReceipt': 'Zobrazit účtenku',
diff --git a/shared/src/i18n/cs/reservations.ts b/shared/src/i18n/cs/reservations.ts
index a474e6c4a..3a9393bad 100644
--- a/shared/src/i18n/cs/reservations.ts
+++ b/shared/src/i18n/cs/reservations.ts
@@ -129,6 +129,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Pusťte soubory pro import',
'reservations.import.acceptedFormats':
'Přijímané formáty: EML, PDF, PKPass, HTML, TXT (max. 10 MB každý, až 5 souborů)',
+ 'reservations.import.acceptedPhotos': 'Fotografie (JPG, PNG, WEBP) čte model AI.',
'reservations.import.parsing': 'Zpracování souborů…',
'reservations.import.previewHeading': 'Nalezeno {count} rezervace/í',
'reservations.import.previewEmpty': 'Z nahraných souborů se nepodařilo extrahovat žádné rezervace.',
diff --git a/shared/src/i18n/cs/settings.ts b/shared/src/i18n/cs/settings.ts
index f2f1fb4a5..8e0199584 100644
--- a/shared/src/i18n/cs/settings.ts
+++ b/shared/src/i18n/cs/settings.ts
@@ -390,9 +390,9 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'Kde model běží — místní server Ollama nebo koncový bod kompatibilní s OpenAI.',
'settings.aiParsing.apiKey': 'Klíč API',
'settings.aiParsing.apiKeyHint': 'Ukládá se šifrovaně. Ponechte prázdné pro zachování aktuálního klíče.',
- 'settings.aiParsing.multimodal': 'Odesílat dokumenty jako obrázky',
+ 'settings.aiParsing.multimodal': 'Model čte obrázky',
'settings.aiParsing.multimodalHint':
- 'Pro modely se schopností zpracovat obraz — odešle původní PDF místo extrahovaného textu.',
+ 'Zapněte pro model, který čte obrázky, aby šlo fotografii importovat nebo naskenovat.',
'settings.aiParsing.toast.saved': 'Nastavení AI uloženo',
'settings.aiParsing.toast.saveError': 'Nastavení AI se nepodařilo uložit',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/de/admin.ts b/shared/src/i18n/de/admin.ts
index d9e1929a0..6beb02405 100644
--- a/shared/src/i18n/de/admin.ts
+++ b/shared/src/i18n/de/admin.ts
@@ -624,6 +624,11 @@ const admin: TranslationStrings = {
'Aufenthalte und aufgezeichnete Routen aus einer Dawarich-Instanz lesen, die jeder selbst verbindet',
'admin.addons.catalog.llm_parsing.name': 'KI-Auswertung',
'admin.addons.catalog.llm_parsing.description': 'Liest Buchungen, an denen der eingebaute Parser scheitert, mit einem KI-Modell deiner Wahl',
+ 'admin.addons.llm.vision.auto': 'Automatisch',
+ 'admin.addons.llm.vision.on': 'Ja',
+ 'admin.addons.llm.vision.off': 'Nein',
+ 'admin.addons.llm.vision.hintLocal': 'Automatisch fragt den Ollama-Server, ob dieses Modell Bilder liest.',
+ 'admin.addons.llm.vision.hintCloud': 'Automatisch bedeutet bei einem Cloud-Modell Nein. Wähle Ja, wenn dieses Modell Bilder liest.',
'admin.addons.enabled': 'Aktiviert',
'admin.addons.disabled': 'Deaktiviert',
'admin.addons.type.trip': 'Reise',
diff --git a/shared/src/i18n/de/budget.ts b/shared/src/i18n/de/budget.ts
index 07c619dea..70f6de170 100644
--- a/shared/src/i18n/de/budget.ts
+++ b/shared/src/i18n/de/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Belege',
'costs.receiptsTitle': 'Belege & Rechnungen',
'costs.attachReceipt': 'Beleg / Rechnung anhängen',
+ 'costs.scan.button': 'Beleg scannen',
+ 'costs.scan.reading': 'Beleg wird gelesen…',
+ 'costs.scan.review': 'Ausgabe prüfen',
+ 'costs.scan.nothingRead': 'Aus diesem Foto ließ sich kein Beleg lesen.',
+ 'costs.scan.failed': 'Der Beleg konnte nicht eingelesen werden.',
+ 'costs.scan.title': 'Beleg scannen',
+ 'costs.scan.accepted': 'Ein Foto eines Belegs (JPG, PNG oder WEBP, bis 10 MB). Das KI-Modell liest es im Hintergrund, danach prüfst du die Ausgabe, bevor sie gespeichert wird.',
+ 'costs.scan.dropHere': 'Belegfoto hier ablegen oder klicken, um eins auszuwählen oder aufzunehmen',
+ 'costs.scan.dropActive': 'Foto ablegen, um es zu scannen',
+ 'costs.scan.start': 'Scannen',
+ 'costs.scan.unsupported': 'Nur Fotos lassen sich scannen: JPG, PNG oder WEBP.',
'costs.noReceipts': 'Keine Belege angehängt',
'costs.deleteReceipt': 'Beleg entfernen',
'costs.viewReceipt': 'Beleg ansehen',
diff --git a/shared/src/i18n/de/reservations.ts b/shared/src/i18n/de/reservations.ts
index a8851f269..052255254 100644
--- a/shared/src/i18n/de/reservations.ts
+++ b/shared/src/i18n/de/reservations.ts
@@ -132,6 +132,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Dateien zum Importieren ablegen',
'reservations.import.acceptedFormats':
'Akzeptiert: EML, PDF, PKPass, HTML, TXT (max. 10 MB pro Datei, bis zu 5 Dateien)',
+ 'reservations.import.acceptedPhotos': 'Fotos (JPG, PNG, WEBP) liest das KI-Modell.',
'reservations.import.parsing': 'Dateien werden verarbeitet…',
'reservations.import.previewHeading': '{count} Reservierung(en) gefunden',
'reservations.import.previewEmpty': 'Aus den hochgeladenen Dateien konnten keine Reservierungen extrahiert werden.',
diff --git a/shared/src/i18n/de/settings.ts b/shared/src/i18n/de/settings.ts
index 7dc650452..c2656d36e 100644
--- a/shared/src/i18n/de/settings.ts
+++ b/shared/src/i18n/de/settings.ts
@@ -401,9 +401,9 @@ const settings: TranslationStrings = {
'Wo das Modell läuft — ein lokaler Ollama-Server oder ein OpenAI-kompatibler Endpunkt.',
'settings.aiParsing.apiKey': 'API-Schlüssel',
'settings.aiParsing.apiKeyHint': 'Verschlüsselt gespeichert. Leer lassen, um den aktuellen Schlüssel zu behalten.',
- 'settings.aiParsing.multimodal': 'Dokumente als Bilder senden',
+ 'settings.aiParsing.multimodal': 'Modell liest Bilder',
'settings.aiParsing.multimodalHint':
- 'Für Modelle mit Bildverständnis — sendet das Original-PDF statt extrahiertem Text.',
+ 'Für ein Modell, das Bilder liest: Dann lässt sich ein Foto importieren oder scannen.',
'settings.aiParsing.toast.saved': 'KI-Einstellungen gespeichert',
'settings.aiParsing.toast.saveError': 'KI-Einstellungen konnten nicht gespeichert werden',
'settings.tabs.appearance': 'Erscheinungsbild',
diff --git a/shared/src/i18n/en/admin.ts b/shared/src/i18n/en/admin.ts
index a38b1bffc..250244066 100644
--- a/shared/src/i18n/en/admin.ts
+++ b/shared/src/i18n/en/admin.ts
@@ -663,6 +663,11 @@ const admin: TranslationStrings = {
'Read visits and recorded routes from a Dawarich instance each reader connects themselves',
'admin.addons.catalog.llm_parsing.name': 'AI Parsing',
'admin.addons.catalog.llm_parsing.description': 'Reads bookings the built-in parser cannot, using an AI model you choose',
+ 'admin.addons.llm.vision.auto': 'Automatic',
+ 'admin.addons.llm.vision.on': 'Yes',
+ 'admin.addons.llm.vision.off': 'No',
+ 'admin.addons.llm.vision.hintLocal': 'Automatic asks the Ollama server whether this model reads images.',
+ 'admin.addons.llm.vision.hintCloud': 'Automatic means no for a cloud model. Choose Yes if this model reads images.',
'admin.addons.enabled': 'Enabled',
'admin.addons.disabled': 'Disabled',
'admin.addons.type.trip': 'Trip',
diff --git a/shared/src/i18n/en/budget.ts b/shared/src/i18n/en/budget.ts
index 8d277c37a..8d2ac6c18 100644
--- a/shared/src/i18n/en/budget.ts
+++ b/shared/src/i18n/en/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Receipts',
'costs.receiptsTitle': 'Receipts & Invoices',
'costs.attachReceipt': 'Attach receipt / invoice',
+ 'costs.scan.button': 'Scan receipt',
+ 'costs.scan.reading': 'Reading the receipt…',
+ 'costs.scan.review': 'Review expense',
+ 'costs.scan.nothingRead': 'No receipt could be read from this photo.',
+ 'costs.scan.failed': 'Could not start reading the receipt.',
+ 'costs.scan.title': 'Scan a receipt',
+ 'costs.scan.accepted': 'A photo of one receipt (JPG, PNG or WEBP, up to 10 MB). The AI model reads it in the background, then you check the expense before it is saved.',
+ 'costs.scan.dropHere': 'Drop the receipt photo here, or click to pick or take one',
+ 'costs.scan.dropActive': 'Drop the photo to scan it',
+ 'costs.scan.start': 'Scan',
+ 'costs.scan.unsupported': 'Only a photo can be scanned: JPG, PNG or WEBP.',
'costs.noReceipts': 'No receipts attached',
'costs.deleteReceipt': 'Remove receipt',
'costs.viewReceipt': 'View receipt',
diff --git a/shared/src/i18n/en/reservations.ts b/shared/src/i18n/en/reservations.ts
index 25e112531..a93b28fbd 100644
--- a/shared/src/i18n/en/reservations.ts
+++ b/shared/src/i18n/en/reservations.ts
@@ -133,6 +133,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Drop booking confirmation files here, or click to select',
'reservations.import.dropActive': 'Drop files to import',
'reservations.import.acceptedFormats': 'Accepted: EML, PDF, PKPass, HTML, TXT (max 10 MB each, up to 5 files)',
+ 'reservations.import.acceptedPhotos': 'Photos (JPG, PNG, WEBP) are read by the AI model.',
'reservations.import.parsing': 'Parsing files…',
'reservations.import.previewHeading': '{count} reservation(s) found',
'reservations.import.previewEmpty': 'No reservations could be extracted from the uploaded files.',
diff --git a/shared/src/i18n/en/settings.ts b/shared/src/i18n/en/settings.ts
index 0b4387e59..f1a3a19b4 100644
--- a/shared/src/i18n/en/settings.ts
+++ b/shared/src/i18n/en/settings.ts
@@ -400,8 +400,8 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'Where the model runs — a local Ollama server or an OpenAI-compatible endpoint.',
'settings.aiParsing.apiKey': 'API key',
'settings.aiParsing.apiKeyHint': 'Stored encrypted. Leave blank to keep the current key.',
- 'settings.aiParsing.multimodal': 'Send documents as images',
- 'settings.aiParsing.multimodalHint': 'For vision-capable models — sends the original PDF instead of extracted text.',
+ 'settings.aiParsing.multimodal': 'Model reads images',
+ 'settings.aiParsing.multimodalHint': 'Turn on for a model that reads images, so a photo can be imported or scanned.',
'settings.aiParsing.toast.saved': 'AI settings saved',
'settings.aiParsing.toast.saveError': 'Could not save AI settings',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/es/admin.ts b/shared/src/i18n/es/admin.ts
index 73e470218..fab59b8a2 100644
--- a/shared/src/i18n/es/admin.ts
+++ b/shared/src/i18n/es/admin.ts
@@ -609,6 +609,11 @@ const admin: TranslationStrings = {
'Lee estancias y rutas grabadas de una instancia de Dawarich que cada usuario conecta por su cuenta',
'admin.addons.catalog.llm_parsing.name': 'Análisis con IA',
'admin.addons.catalog.llm_parsing.description': 'Lee reservas que el analizador integrado no entiende, con un modelo de IA que tú elijas',
+ 'admin.addons.llm.vision.auto': 'Automático',
+ 'admin.addons.llm.vision.on': 'Sí',
+ 'admin.addons.llm.vision.off': 'No',
+ 'admin.addons.llm.vision.hintLocal': 'Automático pregunta al servidor Ollama si este modelo lee imágenes.',
+ 'admin.addons.llm.vision.hintCloud': 'Automático significa no para un modelo en la nube. Elige Sí si este modelo lee imágenes.',
'admin.addons.enabled': 'Activo',
'admin.addons.disabled': 'Desactivado',
'admin.addons.type.trip': 'Viaje',
diff --git a/shared/src/i18n/es/budget.ts b/shared/src/i18n/es/budget.ts
index b92cd8c49..50b46efd7 100644
--- a/shared/src/i18n/es/budget.ts
+++ b/shared/src/i18n/es/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Recibos',
'costs.receiptsTitle': 'Recibos y facturas',
'costs.attachReceipt': 'Adjuntar recibo / factura',
+ 'costs.scan.button': 'Escanear recibo',
+ 'costs.scan.reading': 'Leyendo el recibo…',
+ 'costs.scan.review': 'Revisar el gasto',
+ 'costs.scan.nothingRead': 'No se pudo leer ningún recibo en esta foto.',
+ 'costs.scan.failed': 'No se pudo iniciar la lectura del recibo.',
+ 'costs.scan.title': 'Escanear recibo',
+ 'costs.scan.accepted': 'Una foto de un solo recibo (JPG, PNG o WEBP, hasta 10 MB). El modelo de IA la lee en segundo plano y luego revisas el gasto antes de guardarlo.',
+ 'costs.scan.dropHere': 'Suelta aquí la foto del recibo, o haz clic para elegir o hacer una',
+ 'costs.scan.dropActive': 'Suelta la foto para escanearla',
+ 'costs.scan.start': 'Escanear',
+ 'costs.scan.unsupported': 'Solo se puede escanear una foto: JPG, PNG o WEBP.',
'costs.noReceipts': 'No hay recibos adjuntos',
'costs.deleteReceipt': 'Eliminar recibo',
'costs.viewReceipt': 'Ver recibo',
diff --git a/shared/src/i18n/es/reservations.ts b/shared/src/i18n/es/reservations.ts
index a104831b9..e20672f5d 100644
--- a/shared/src/i18n/es/reservations.ts
+++ b/shared/src/i18n/es/reservations.ts
@@ -131,6 +131,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Suelta los archivos para importar',
'reservations.import.acceptedFormats':
'Aceptados: EML, PDF, PKPass, HTML, TXT (máx. 10 MB por archivo, hasta 5 archivos)',
+ 'reservations.import.acceptedPhotos': 'El modelo de IA lee las fotos (JPG, PNG, WEBP).',
'reservations.import.parsing': 'Analizando archivos…',
'reservations.import.previewHeading': '{count} reserva(s) encontrada(s)',
'reservations.import.previewEmpty': 'No se pudieron extraer reservas de los archivos subidos.',
diff --git a/shared/src/i18n/es/settings.ts b/shared/src/i18n/es/settings.ts
index e612ae95e..53a5f8a77 100644
--- a/shared/src/i18n/es/settings.ts
+++ b/shared/src/i18n/es/settings.ts
@@ -399,8 +399,8 @@ const settings: TranslationStrings = {
'Donde se ejecuta el modelo: un servidor Ollama local o un endpoint compatible con OpenAI.',
'settings.aiParsing.apiKey': 'Clave de API',
'settings.aiParsing.apiKeyHint': 'Se almacena cifrada. Déjalo en blanco para mantener la clave actual.',
- 'settings.aiParsing.multimodal': 'Enviar documentos como imágenes',
- 'settings.aiParsing.multimodalHint': 'Para modelos con visión: envía el PDF original en lugar del texto extraído.',
+ 'settings.aiParsing.multimodal': 'El modelo lee imágenes',
+ 'settings.aiParsing.multimodalHint': 'Actívalo para un modelo que lee imágenes, así se puede importar o escanear una foto.',
'settings.aiParsing.toast.saved': 'Ajustes de IA guardados',
'settings.aiParsing.toast.saveError': 'No se han podido guardar los ajustes de IA',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/et/admin.ts b/shared/src/i18n/et/admin.ts
index 6c2a0a5a0..09ab386e5 100644
--- a/shared/src/i18n/et/admin.ts
+++ b/shared/src/i18n/et/admin.ts
@@ -681,6 +681,11 @@ const admin: TranslationStrings = {
'admin.addons.catalog.llm_parsing.name': 'Tehisintellektiga parsimine',
'admin.addons.catalog.llm_parsing.description':
'Loeb sinu valitud tehisintellektimudeli abil broneeringuid, mida sisseehitatud parser lugeda ei suuda',
+ 'admin.addons.llm.vision.auto': 'Automaatne',
+ 'admin.addons.llm.vision.on': 'Jah',
+ 'admin.addons.llm.vision.off': 'Ei',
+ 'admin.addons.llm.vision.hintLocal': 'Automaatne küsib Ollama serverilt, kas see mudel loeb pilte.',
+ 'admin.addons.llm.vision.hintCloud': 'Pilvemudeli puhul tähendab Automaatne „ei”. Vali Jah, kui see mudel loeb pilte.',
'admin.addons.enabled': 'Lubatud',
'admin.addons.disabled': 'Keelatud',
'admin.addons.type.trip': 'Reis',
diff --git a/shared/src/i18n/et/budget.ts b/shared/src/i18n/et/budget.ts
index 2e4643251..8a8435559 100644
--- a/shared/src/i18n/et/budget.ts
+++ b/shared/src/i18n/et/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Kviitungid',
'costs.receiptsTitle': 'Kviitungid ja arved',
'costs.attachReceipt': 'Lisa kviitung või arve',
+ 'costs.scan.button': 'Skanni kviitung',
+ 'costs.scan.reading': 'Kviitungit loetakse…',
+ 'costs.scan.review': 'Vaata kulu üle',
+ 'costs.scan.nothingRead': 'Sellelt fotolt ei õnnestunud kviitungit lugeda.',
+ 'costs.scan.failed': 'Kviitungi lugemist ei õnnestunud alustada.',
+ 'costs.scan.title': 'Skanni kviitung',
+ 'costs.scan.accepted': 'Ühe kviitungi foto (JPG, PNG või WEBP, kuni 10 MB). Tehisintellekti mudel loeb selle taustal, seejärel kontrollid kulu enne salvestamist.',
+ 'costs.scan.dropHere': 'Lohista kviitungi foto siia või klõpsa, et valida või pildistada',
+ 'costs.scan.dropActive': 'Lase foto lahti, et see skannida',
+ 'costs.scan.start': 'Skanni',
+ 'costs.scan.unsupported': 'Skannida saab ainult fotot: JPG, PNG või WEBP.',
'costs.noReceipts': 'Kviitungeid pole lisatud',
'costs.deleteReceipt': 'Eemalda kviitung',
'costs.viewReceipt': 'Vaata kviitungit',
diff --git a/shared/src/i18n/et/reservations.ts b/shared/src/i18n/et/reservations.ts
index d747c4787..76fae685b 100644
--- a/shared/src/i18n/et/reservations.ts
+++ b/shared/src/i18n/et/reservations.ts
@@ -133,6 +133,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Lohista broneeringukinnituse failid siia või klõpsa valimiseks',
'reservations.import.dropActive': 'Impordiks lohista failid siia',
'reservations.import.acceptedFormats': 'Lubatud: EML, PDF, PKPass, HTML, TXT (kuni 10 MB faili kohta, kuni 5 faili)',
+ 'reservations.import.acceptedPhotos': 'Fotosid (JPG, PNG, WEBP) loeb tehisintellekti mudel.',
'reservations.import.parsing': 'Failide töötlemine…',
'reservations.import.previewHeading': 'Leiti {count} broneeringut',
'reservations.import.previewEmpty': 'Üleslaaditud failidest ei õnnestunud broneeringuid eraldada.',
diff --git a/shared/src/i18n/et/settings.ts b/shared/src/i18n/et/settings.ts
index 1e2041438..a4849dcb9 100644
--- a/shared/src/i18n/et/settings.ts
+++ b/shared/src/i18n/et/settings.ts
@@ -405,8 +405,8 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'Mudeli asukoht — kohalik Ollama server või OpenAI-ühilduv otspunkt.',
'settings.aiParsing.apiKey': 'API-võti',
'settings.aiParsing.apiKeyHint': 'Salvestatakse krüptitult. Praeguse võtme säilitamiseks jäta tühjaks.',
- 'settings.aiParsing.multimodal': 'Saada dokumendid piltidena',
- 'settings.aiParsing.multimodalHint': 'Pildituvastusega mudelitele — saadab eraldatud teksti asemel algse PDF-i.',
+ 'settings.aiParsing.multimodal': 'Mudel loeb pilte',
+ 'settings.aiParsing.multimodalHint': 'Lülita sisse pilte lugeva mudeli jaoks, siis saab foto importida või skannida.',
'settings.aiParsing.toast.saved': 'Tehisintellekti seaded salvestatud',
'settings.aiParsing.toast.saveError': 'Tehisintellekti seadete salvestamine ebaõnnestus',
'settings.tabs.appearance': 'Välimus',
diff --git a/shared/src/i18n/fr/admin.ts b/shared/src/i18n/fr/admin.ts
index 202752248..9a27e14c3 100644
--- a/shared/src/i18n/fr/admin.ts
+++ b/shared/src/i18n/fr/admin.ts
@@ -630,6 +630,11 @@ const admin: TranslationStrings = {
'Lisez les séjours et les trajets enregistrés depuis une instance Dawarich que chaque utilisateur connecte lui-même',
'admin.addons.catalog.llm_parsing.name': 'Analyse par IA',
'admin.addons.catalog.llm_parsing.description': 'Lit les réservations que l’analyseur intégré ne comprend pas, via un modèle d’IA de votre choix',
+ 'admin.addons.llm.vision.auto': 'Automatique',
+ 'admin.addons.llm.vision.on': 'Oui',
+ 'admin.addons.llm.vision.off': 'Non',
+ 'admin.addons.llm.vision.hintLocal': 'En automatique, TREK demande au serveur Ollama si ce modèle lit les images.',
+ 'admin.addons.llm.vision.hintCloud': 'En automatique, un modèle cloud est considéré comme ne lisant pas les images. Choisissez Oui si ce modèle les lit.',
'admin.addons.enabled': 'Activé',
'admin.addons.disabled': 'Désactivé',
'admin.addons.type.trip': 'Voyage',
diff --git a/shared/src/i18n/fr/budget.ts b/shared/src/i18n/fr/budget.ts
index c2b054160..20dc5d21f 100644
--- a/shared/src/i18n/fr/budget.ts
+++ b/shared/src/i18n/fr/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Reçus',
'costs.receiptsTitle': 'Reçus et factures',
'costs.attachReceipt': 'Joindre un reçu / une facture',
+ 'costs.scan.button': 'Scanner un reçu',
+ 'costs.scan.reading': 'Lecture du reçu…',
+ 'costs.scan.review': 'Vérifier la dépense',
+ 'costs.scan.nothingRead': "Aucun reçu n'a pu être lu sur cette photo.",
+ 'costs.scan.failed': 'Impossible de lancer la lecture du reçu.',
+ 'costs.scan.title': 'Scanner un reçu',
+ 'costs.scan.accepted': "Une photo d'un seul reçu (JPG, PNG ou WEBP, 10 Mo max). Le modèle d'IA la lit en arrière-plan, puis vous vérifiez la dépense avant de l'enregistrer.",
+ 'costs.scan.dropHere': 'Déposez la photo du reçu ici, ou cliquez pour en choisir ou en prendre une',
+ 'costs.scan.dropActive': 'Déposez la photo pour la scanner',
+ 'costs.scan.start': 'Scanner',
+ 'costs.scan.unsupported': 'Seule une photo peut être scannée : JPG, PNG ou WEBP.',
'costs.noReceipts': 'Aucun reçu joint',
'costs.deleteReceipt': 'Supprimer le reçu',
'costs.viewReceipt': 'Voir le reçu',
diff --git a/shared/src/i18n/fr/reservations.ts b/shared/src/i18n/fr/reservations.ts
index 9444974de..a07d0dd43 100644
--- a/shared/src/i18n/fr/reservations.ts
+++ b/shared/src/i18n/fr/reservations.ts
@@ -132,6 +132,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Déposez les fichiers pour importer',
'reservations.import.acceptedFormats':
"Acceptés : EML, PDF, PKPass, HTML, TXT (max. 10 Mo chacun, jusqu'à 5 fichiers)",
+ 'reservations.import.acceptedPhotos': "Les photos (JPG, PNG, WEBP) sont lues par le modèle d'IA.",
'reservations.import.parsing': 'Analyse des fichiers…',
'reservations.import.previewHeading': '{count} réservation(s) trouvée(s)',
'reservations.import.previewEmpty': "Aucune réservation n'a pu être extraite des fichiers envoyés.",
diff --git a/shared/src/i18n/fr/settings.ts b/shared/src/i18n/fr/settings.ts
index 8c40b91a6..77c068eb1 100644
--- a/shared/src/i18n/fr/settings.ts
+++ b/shared/src/i18n/fr/settings.ts
@@ -406,9 +406,9 @@ const settings: TranslationStrings = {
"Emplacement d'exécution du modèle — un serveur Ollama local ou un point de terminaison compatible OpenAI.",
'settings.aiParsing.apiKey': 'Clé API',
'settings.aiParsing.apiKeyHint': 'Stockée de façon chiffrée. Laissez vide pour conserver la clé actuelle.',
- 'settings.aiParsing.multimodal': "Envoyer les documents sous forme d'images",
+ 'settings.aiParsing.multimodal': 'Le modèle lit les images',
'settings.aiParsing.multimodalHint':
- "Pour les modèles capables d'analyser des images — envoie le PDF d'origine au lieu du texte extrait.",
+ 'À activer pour un modèle qui lit les images : une photo peut alors être importée ou scannée.',
'settings.aiParsing.toast.saved': "Paramètres d'IA enregistrés",
'settings.aiParsing.toast.saveError': "Impossible d'enregistrer les paramètres d'IA",
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/gr/admin.ts b/shared/src/i18n/gr/admin.ts
index 0413054ec..f54ebad0a 100644
--- a/shared/src/i18n/gr/admin.ts
+++ b/shared/src/i18n/gr/admin.ts
@@ -685,6 +685,11 @@ const admin: TranslationStrings = {
'Ανάγνωση επισκέψεων και καταγεγραμμένων διαδρομών από μια εγκατάσταση Dawarich που συνδέει ο κάθε χρήστης μόνος του',
'admin.addons.catalog.llm_parsing.name': 'Ανάλυση με τεχνητή νοημοσύνη',
'admin.addons.catalog.llm_parsing.description': 'Διαβάζει κρατήσεις που δεν καταλαβαίνει ο ενσωματωμένος αναλυτής, με μοντέλο ΤΝ της επιλογής σας',
+ 'admin.addons.llm.vision.auto': 'Αυτόματα',
+ 'admin.addons.llm.vision.on': 'Ναι',
+ 'admin.addons.llm.vision.off': 'Όχι',
+ 'admin.addons.llm.vision.hintLocal': 'Η επιλογή Αυτόματα ρωτά τον διακομιστή Ollama αν αυτό το μοντέλο διαβάζει εικόνες.',
+ 'admin.addons.llm.vision.hintCloud': 'Για μοντέλο στο cloud, η επιλογή Αυτόματα σημαίνει όχι. Επιλέξτε Ναι αν αυτό το μοντέλο διαβάζει εικόνες.',
'admin.addons.enabled': 'Ενεργοποιημένο',
'admin.addons.disabled': 'Απενεργοποιημένο',
'admin.addons.type.trip': 'Ταξίδι',
diff --git a/shared/src/i18n/gr/budget.ts b/shared/src/i18n/gr/budget.ts
index 66ac7d95c..80b8a2baf 100644
--- a/shared/src/i18n/gr/budget.ts
+++ b/shared/src/i18n/gr/budget.ts
@@ -155,6 +155,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Αποδείξεις',
'costs.receiptsTitle': 'Αποδείξεις & Τιμολόγια',
'costs.attachReceipt': 'Επισύναψη απόδειξης / τιμολογίου',
+ 'costs.scan.button': 'Σάρωση απόδειξης',
+ 'costs.scan.reading': 'Ανάγνωση της απόδειξης…',
+ 'costs.scan.review': 'Έλεγχος εξόδου',
+ 'costs.scan.nothingRead': 'Δεν ήταν δυνατή η ανάγνωση απόδειξης από αυτή τη φωτογραφία.',
+ 'costs.scan.failed': 'Δεν ήταν δυνατή η έναρξη ανάγνωσης της απόδειξης.',
+ 'costs.scan.title': 'Σάρωση απόδειξης',
+ 'costs.scan.accepted': 'Φωτογραφία μίας απόδειξης (JPG, PNG ή WEBP, έως 10 MB). Το μοντέλο ΤΝ τη διαβάζει στο παρασκήνιο και μετά ελέγχετε το έξοδο πριν αποθηκευτεί.',
+ 'costs.scan.dropHere': 'Αφήστε εδώ τη φωτογραφία της απόδειξης ή κάντε κλικ για να επιλέξετε ή να τραβήξετε μία',
+ 'costs.scan.dropActive': 'Αφήστε τη φωτογραφία για σάρωση',
+ 'costs.scan.start': 'Σάρωση',
+ 'costs.scan.unsupported': 'Μόνο φωτογραφία μπορεί να σαρωθεί: JPG, PNG ή WEBP.',
'costs.noReceipts': 'Δεν υπάρχουν συνημμένες αποδείξεις',
'costs.deleteReceipt': 'Αφαίρεση απόδειξης',
'costs.viewReceipt': 'Προβολή απόδειξης',
diff --git a/shared/src/i18n/gr/reservations.ts b/shared/src/i18n/gr/reservations.ts
index b916556cd..e71270785 100644
--- a/shared/src/i18n/gr/reservations.ts
+++ b/shared/src/i18n/gr/reservations.ts
@@ -131,6 +131,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Αποθέστε αρχεία για εισαγωγή',
'reservations.import.acceptedFormats':
'Αποδεκτά: EML, PDF, PKPass, HTML, TXT (μέγιστο 10 MB το καθένα, έως 5 αρχεία)',
+ 'reservations.import.acceptedPhotos': 'Τις φωτογραφίες (JPG, PNG, WEBP) τις διαβάζει το μοντέλο ΤΝ.',
'reservations.import.parsing': 'Επεξεργασία αρχείων…',
'reservations.import.previewHeading': 'Βρέθηκαν {count} κράτηση/κρατήσεις',
'reservations.import.previewEmpty': 'Δεν ήταν δυνατή η εξαγωγή κρατήσεων από τα μεταφορτωμένα αρχεία.',
diff --git a/shared/src/i18n/gr/settings.ts b/shared/src/i18n/gr/settings.ts
index e4b5e55b2..464a6a34a 100644
--- a/shared/src/i18n/gr/settings.ts
+++ b/shared/src/i18n/gr/settings.ts
@@ -409,9 +409,9 @@ const settings: TranslationStrings = {
'Πού εκτελείται το μοντέλο — ένας τοπικός διακομιστής Ollama ή ένα τελικό σημείο συμβατό με OpenAI.',
'settings.aiParsing.apiKey': 'Κλειδί API',
'settings.aiParsing.apiKeyHint': 'Αποθηκεύεται κρυπτογραφημένο. Αφήστε το κενό για να διατηρήσετε το τρέχον κλειδί.',
- 'settings.aiParsing.multimodal': 'Αποστολή εγγράφων ως εικόνες',
+ 'settings.aiParsing.multimodal': 'Το μοντέλο διαβάζει εικόνες',
'settings.aiParsing.multimodalHint':
- 'Για μοντέλα με δυνατότητα όρασης — στέλνει το αρχικό PDF αντί για το εξαγόμενο κείμενο.',
+ 'Ενεργοποιήστε το για μοντέλο που διαβάζει εικόνες, ώστε μια φωτογραφία να μπορεί να εισαχθεί ή να σαρωθεί.',
'settings.aiParsing.toast.saved': 'Οι ρυθμίσεις AI αποθηκεύτηκαν',
'settings.aiParsing.toast.saveError': 'Δεν ήταν δυνατή η αποθήκευση των ρυθμίσεων AI',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/hu/admin.ts b/shared/src/i18n/hu/admin.ts
index b4dc623b7..16db47374 100644
--- a/shared/src/i18n/hu/admin.ts
+++ b/shared/src/i18n/hu/admin.ts
@@ -630,6 +630,11 @@ const admin: TranslationStrings = {
'Tartózkodások és rögzített útvonalak olvasása egy Dawarich-példányból, amelyet minden olvasó maga köt be',
'admin.addons.catalog.llm_parsing.name': 'MI-elemzés',
'admin.addons.catalog.llm_parsing.description': 'Beolvassa azokat a foglalásokat, amelyeket a beépített értelmező nem tud, az általad választott MI-modellel',
+ 'admin.addons.llm.vision.auto': 'Automatikus',
+ 'admin.addons.llm.vision.on': 'Igen',
+ 'admin.addons.llm.vision.off': 'Nem',
+ 'admin.addons.llm.vision.hintLocal': 'Az Automatikus megkérdezi az Ollama-szervert, hogy ez a modell olvas-e képeket.',
+ 'admin.addons.llm.vision.hintCloud': 'Felhőalapú modellnél az Automatikus nemet jelent. Válaszd az Igent, ha ez a modell olvas képeket.',
'admin.addons.enabled': 'Engedélyezve',
'admin.addons.disabled': 'Letiltva',
'admin.addons.type.trip': 'Utazás',
diff --git a/shared/src/i18n/hu/budget.ts b/shared/src/i18n/hu/budget.ts
index 8354424e1..51b608d07 100644
--- a/shared/src/i18n/hu/budget.ts
+++ b/shared/src/i18n/hu/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Nyugták',
'costs.receiptsTitle': 'Nyugták és számlák',
'costs.attachReceipt': 'Nyugta / számla csatolása',
+ 'costs.scan.button': 'Nyugta beolvasása',
+ 'costs.scan.reading': 'Nyugta olvasása…',
+ 'costs.scan.review': 'Kiadás ellenőrzése',
+ 'costs.scan.nothingRead': 'Erről a fotóról nem sikerült nyugtát olvasni.',
+ 'costs.scan.failed': 'Nem sikerült elindítani a nyugta olvasását.',
+ 'costs.scan.title': 'Nyugta beolvasása',
+ 'costs.scan.accepted': 'Egyetlen nyugta fotója (JPG, PNG vagy WEBP, legfeljebb 10 MB). Az MI-modell a háttérben olvassa be, majd mentés előtt ellenőrzöd a kiadást.',
+ 'costs.scan.dropHere': 'Húzd ide a nyugta fotóját, vagy kattints egy kiválasztásához vagy elkészítéséhez',
+ 'costs.scan.dropActive': 'Engedd el a fotót a beolvasáshoz',
+ 'costs.scan.start': 'Beolvasás',
+ 'costs.scan.unsupported': 'Csak fotó olvasható be: JPG, PNG vagy WEBP.',
'costs.noReceipts': 'Nincsenek csatolt nyugták',
'costs.deleteReceipt': 'Nyugta eltávolítása',
'costs.viewReceipt': 'Nyugta megtekintése',
diff --git a/shared/src/i18n/hu/reservations.ts b/shared/src/i18n/hu/reservations.ts
index b58dd4651..23bd8f91e 100644
--- a/shared/src/i18n/hu/reservations.ts
+++ b/shared/src/i18n/hu/reservations.ts
@@ -130,6 +130,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Dobja ide a fájlokat az importáláshoz',
'reservations.import.acceptedFormats':
'Elfogadott: EML, PDF, PKPass, HTML, TXT (max. 10 MB darabonként, legfeljebb 5 fájl)',
+ 'reservations.import.acceptedPhotos': 'A fotókat (JPG, PNG, WEBP) az MI-modell olvassa be.',
'reservations.import.parsing': 'Fájlok feldolgozása…',
'reservations.import.previewHeading': '{count} foglalás találva',
'reservations.import.previewEmpty': 'A feltöltött fájlokból nem sikerült foglalásokat kinyerni.',
diff --git a/shared/src/i18n/hu/settings.ts b/shared/src/i18n/hu/settings.ts
index 13264026c..b5867f5e2 100644
--- a/shared/src/i18n/hu/settings.ts
+++ b/shared/src/i18n/hu/settings.ts
@@ -398,9 +398,9 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'Ahol a modell fut — helyi Ollama-kiszolgáló vagy OpenAI-kompatibilis végpont.',
'settings.aiParsing.apiKey': 'API-kulcs',
'settings.aiParsing.apiKeyHint': 'Titkosítva tárolva. Hagyd üresen a jelenlegi kulcs megtartásához.',
- 'settings.aiParsing.multimodal': 'Dokumentumok küldése képként',
+ 'settings.aiParsing.multimodal': 'A modell olvas képeket',
'settings.aiParsing.multimodalHint':
- 'Képfelismerésre képes modellekhez — az eredeti PDF-et küldi a kinyert szöveg helyett.',
+ 'Kapcsold be képeket olvasó modellhez, így egy fotó importálható vagy beolvasható.',
'settings.aiParsing.toast.saved': 'AI-beállítások elmentve',
'settings.aiParsing.toast.saveError': 'Az AI-beállítások mentése nem sikerült',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/id/admin.ts b/shared/src/i18n/id/admin.ts
index 13df8884f..f3ffd973c 100644
--- a/shared/src/i18n/id/admin.ts
+++ b/shared/src/i18n/id/admin.ts
@@ -674,6 +674,11 @@ const admin: TranslationStrings = {
'Baca kunjungan dan rute terekam dari instans Dawarich yang dihubungkan sendiri oleh tiap pembaca',
'admin.addons.catalog.llm_parsing.name': 'Analisis AI',
'admin.addons.catalog.llm_parsing.description': 'Membaca pemesanan yang tidak dipahami pengurai bawaan, dengan model AI pilihanmu',
+ 'admin.addons.llm.vision.auto': 'Otomatis',
+ 'admin.addons.llm.vision.on': 'Ya',
+ 'admin.addons.llm.vision.off': 'Tidak',
+ 'admin.addons.llm.vision.hintLocal': 'Otomatis menanyakan ke server Ollama apakah model ini membaca gambar.',
+ 'admin.addons.llm.vision.hintCloud': 'Otomatis berarti tidak untuk model cloud. Pilih Ya jika model ini membaca gambar.',
'admin.addons.enabled': 'Aktif',
'admin.addons.disabled': 'Nonaktif',
'admin.addons.type.trip': 'Perjalanan',
diff --git a/shared/src/i18n/id/budget.ts b/shared/src/i18n/id/budget.ts
index 8a27bc183..8ceb8969e 100644
--- a/shared/src/i18n/id/budget.ts
+++ b/shared/src/i18n/id/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Kuitansi',
'costs.receiptsTitle': 'Kuitansi & Faktur',
'costs.attachReceipt': 'Lampirkan kuitansi / faktur',
+ 'costs.scan.button': 'Pindai kuitansi',
+ 'costs.scan.reading': 'Membaca kuitansi…',
+ 'costs.scan.review': 'Tinjau pengeluaran',
+ 'costs.scan.nothingRead': 'Tidak ada kuitansi yang dapat dibaca dari foto ini.',
+ 'costs.scan.failed': 'Tidak dapat mulai membaca kuitansi.',
+ 'costs.scan.title': 'Pindai kuitansi',
+ 'costs.scan.accepted': 'Foto satu kuitansi (JPG, PNG, atau WEBP, maks. 10 MB). Model AI membacanya di latar belakang, lalu kamu memeriksa pengeluaran sebelum disimpan.',
+ 'costs.scan.dropHere': 'Letakkan foto kuitansi di sini, atau klik untuk memilih atau mengambil foto',
+ 'costs.scan.dropActive': 'Lepaskan foto untuk memindainya',
+ 'costs.scan.start': 'Pindai',
+ 'costs.scan.unsupported': 'Hanya foto yang dapat dipindai: JPG, PNG, atau WEBP.',
'costs.noReceipts': 'Tidak ada kuitansi terlampir',
'costs.deleteReceipt': 'Hapus kuitansi',
'costs.viewReceipt': 'Lihat kuitansi',
diff --git a/shared/src/i18n/id/reservations.ts b/shared/src/i18n/id/reservations.ts
index 3757546f5..8484409f3 100644
--- a/shared/src/i18n/id/reservations.ts
+++ b/shared/src/i18n/id/reservations.ts
@@ -129,6 +129,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Seret file konfirmasi pemesanan ke sini atau klik untuk memilih',
'reservations.import.dropActive': 'Lepaskan file untuk mengimpor',
'reservations.import.acceptedFormats': 'Diterima: EML, PDF, PKPass, HTML, TXT (maks. 10 MB per file, hingga 5 file)',
+ 'reservations.import.acceptedPhotos': 'Foto (JPG, PNG, WEBP) dibaca oleh model AI.',
'reservations.import.parsing': 'Memproses file…',
'reservations.import.previewHeading': '{count} pemesanan ditemukan',
'reservations.import.previewEmpty': 'Tidak ada pemesanan yang dapat diekstrak dari file yang diunggah.',
diff --git a/shared/src/i18n/id/settings.ts b/shared/src/i18n/id/settings.ts
index 551542a66..4751290fd 100644
--- a/shared/src/i18n/id/settings.ts
+++ b/shared/src/i18n/id/settings.ts
@@ -396,9 +396,9 @@ const settings: TranslationStrings = {
'Tempat model berjalan — server Ollama lokal atau endpoint yang kompatibel dengan OpenAI.',
'settings.aiParsing.apiKey': 'Kunci API',
'settings.aiParsing.apiKeyHint': 'Disimpan secara terenkripsi. Biarkan kosong untuk mempertahankan kunci saat ini.',
- 'settings.aiParsing.multimodal': 'Kirim dokumen sebagai gambar',
+ 'settings.aiParsing.multimodal': 'Model membaca gambar',
'settings.aiParsing.multimodalHint':
- 'Untuk model yang mendukung visi — mengirim PDF asli alih-alih teks yang diekstrak.',
+ 'Aktifkan untuk model yang membaca gambar, agar foto dapat diimpor atau dipindai.',
'settings.aiParsing.toast.saved': 'Pengaturan AI disimpan',
'settings.aiParsing.toast.saveError': 'Tidak dapat menyimpan pengaturan AI',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/it/admin.ts b/shared/src/i18n/it/admin.ts
index 9377866a0..19a80ef1c 100644
--- a/shared/src/i18n/it/admin.ts
+++ b/shared/src/i18n/it/admin.ts
@@ -625,6 +625,11 @@ const admin: TranslationStrings = {
'Legge le soste e i percorsi registrati da un’istanza Dawarich che ogni utente collega da sé',
'admin.addons.catalog.llm_parsing.name': 'Analisi con IA',
'admin.addons.catalog.llm_parsing.description': 'Legge le prenotazioni che l’analizzatore integrato non riesce a interpretare, con un modello di IA a tua scelta',
+ 'admin.addons.llm.vision.auto': 'Automatico',
+ 'admin.addons.llm.vision.on': 'Sì',
+ 'admin.addons.llm.vision.off': 'No',
+ 'admin.addons.llm.vision.hintLocal': 'Automatico chiede al server Ollama se questo modello legge le immagini.',
+ 'admin.addons.llm.vision.hintCloud': 'Automatico significa no per un modello cloud. Scegli Sì se questo modello legge le immagini.',
'admin.addons.enabled': 'Abilitato',
'admin.addons.disabled': 'Disabilitato',
'admin.addons.type.trip': 'Viaggio',
diff --git a/shared/src/i18n/it/budget.ts b/shared/src/i18n/it/budget.ts
index bce30f879..ef6eaf00e 100644
--- a/shared/src/i18n/it/budget.ts
+++ b/shared/src/i18n/it/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Ricevute',
'costs.receiptsTitle': 'Ricevute e fatture',
'costs.attachReceipt': 'Allega ricevuta / fattura',
+ 'costs.scan.button': 'Scansiona ricevuta',
+ 'costs.scan.reading': 'Lettura della ricevuta…',
+ 'costs.scan.review': 'Controlla la spesa',
+ 'costs.scan.nothingRead': 'Non è stato possibile leggere una ricevuta da questa foto.',
+ 'costs.scan.failed': 'Impossibile avviare la lettura della ricevuta.',
+ 'costs.scan.title': 'Scansiona ricevuta',
+ 'costs.scan.accepted': 'La foto di una sola ricevuta (JPG, PNG o WEBP, fino a 10 MB). Il modello di IA la legge in background, poi controlli la spesa prima di salvarla.',
+ 'costs.scan.dropHere': 'Trascina qui la foto della ricevuta, oppure fai clic per sceglierla o scattarla',
+ 'costs.scan.dropActive': 'Rilascia la foto per scansionarla',
+ 'costs.scan.start': 'Scansiona',
+ 'costs.scan.unsupported': 'Si può scansionare solo una foto: JPG, PNG o WEBP.',
'costs.noReceipts': 'Nessuna ricevuta allegata',
'costs.deleteReceipt': 'Rimuovi ricevuta',
'costs.viewReceipt': 'Visualizza ricevuta',
diff --git a/shared/src/i18n/it/reservations.ts b/shared/src/i18n/it/reservations.ts
index 403577beb..a3714f9a2 100644
--- a/shared/src/i18n/it/reservations.ts
+++ b/shared/src/i18n/it/reservations.ts
@@ -129,6 +129,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Trascina i file di conferma prenotazione qui o clicca per selezionare',
'reservations.import.dropActive': 'Rilascia i file per importare',
'reservations.import.acceptedFormats': 'Accettati: EML, PDF, PKPass, HTML, TXT (max 10 MB ciascuno, fino a 5 file)',
+ 'reservations.import.acceptedPhotos': 'Le foto (JPG, PNG, WEBP) vengono lette dal modello di IA.',
'reservations.import.parsing': 'Analisi dei file in corso…',
'reservations.import.previewHeading': '{count} prenotazione/i trovata/e',
'reservations.import.previewEmpty': 'Nessuna prenotazione è stata estratta dai file caricati.',
diff --git a/shared/src/i18n/it/settings.ts b/shared/src/i18n/it/settings.ts
index ef92cd4cf..d35889724 100644
--- a/shared/src/i18n/it/settings.ts
+++ b/shared/src/i18n/it/settings.ts
@@ -398,9 +398,9 @@ const settings: TranslationStrings = {
'Dove gira il modello: un server Ollama locale o un endpoint compatibile con OpenAI.',
'settings.aiParsing.apiKey': 'Chiave API',
'settings.aiParsing.apiKeyHint': 'Memorizzata in forma cifrata. Lascia vuoto per mantenere la chiave attuale.',
- 'settings.aiParsing.multimodal': 'Invia i documenti come immagini',
+ 'settings.aiParsing.multimodal': 'Il modello legge le immagini',
'settings.aiParsing.multimodalHint':
- 'Per i modelli con capacità visive: invia il PDF originale invece del testo estratto.',
+ 'Attiva per un modello che legge le immagini, così una foto può essere importata o scansionata.',
'settings.aiParsing.toast.saved': 'Impostazioni AI salvate',
'settings.aiParsing.toast.saveError': 'Impossibile salvare le impostazioni AI',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/ja/admin.ts b/shared/src/i18n/ja/admin.ts
index f66f513da..43612bb49 100644
--- a/shared/src/i18n/ja/admin.ts
+++ b/shared/src/i18n/ja/admin.ts
@@ -638,6 +638,11 @@ const admin: TranslationStrings = {
'各自が接続した Dawarich インスタンスから、滞在と記録されたルートを読み取ります',
'admin.addons.catalog.llm_parsing.name': 'AI 解析',
'admin.addons.catalog.llm_parsing.description': '内蔵パーサーが読めない予約を、選んだ AI モデルで読み取ります',
+ 'admin.addons.llm.vision.auto': '自動',
+ 'admin.addons.llm.vision.on': 'はい',
+ 'admin.addons.llm.vision.off': 'いいえ',
+ 'admin.addons.llm.vision.hintLocal': '「自動」では、このモデルが画像を読めるかを Ollama サーバーに問い合わせます。',
+ 'admin.addons.llm.vision.hintCloud': 'クラウドのモデルでは「自動」は「いいえ」になります。このモデルが画像を読める場合は「はい」を選んでください。',
'admin.addons.enabled': '有効',
'admin.addons.disabled': '無効',
'admin.addons.type.trip': '旅行',
diff --git a/shared/src/i18n/ja/budget.ts b/shared/src/i18n/ja/budget.ts
index c1d58ce95..82aa428b3 100644
--- a/shared/src/i18n/ja/budget.ts
+++ b/shared/src/i18n/ja/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'レシート',
'costs.receiptsTitle': 'レシート・請求書',
'costs.attachReceipt': 'レシート・請求書を添付',
+ 'costs.scan.button': 'レシートをスキャン',
+ 'costs.scan.reading': 'レシートを読み取り中…',
+ 'costs.scan.review': '支出を確認',
+ 'costs.scan.nothingRead': 'この写真からレシートを読み取れませんでした。',
+ 'costs.scan.failed': 'レシートの読み取りを開始できませんでした。',
+ 'costs.scan.title': 'レシートをスキャン',
+ 'costs.scan.accepted': 'レシート1枚の写真（JPG、PNG、WEBP、10 MB まで）。AI モデルがバックグラウンドで読み取り、保存前に支出を確認します。',
+ 'costs.scan.dropHere': 'ここにレシートの写真をドロップするか、クリックして選択または撮影',
+ 'costs.scan.dropActive': 'ドロップしてスキャン',
+ 'costs.scan.start': 'スキャン',
+ 'costs.scan.unsupported': 'スキャンできるのは写真（JPG、PNG、WEBP）だけです。',
'costs.noReceipts': '添付されたレシートはありません',
'costs.deleteReceipt': 'レシートを削除',
'costs.viewReceipt': 'レシートを表示',
diff --git a/shared/src/i18n/ja/reservations.ts b/shared/src/i18n/ja/reservations.ts
index e03247da0..1be087cd8 100644
--- a/shared/src/i18n/ja/reservations.ts
+++ b/shared/src/i18n/ja/reservations.ts
@@ -128,6 +128,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': '予約確認ファイルをここにドロップするか、クリックして選択',
'reservations.import.dropActive': 'ファイルをドロップしてインポート',
'reservations.import.acceptedFormats': '対応形式：EML、PDF、PKPass、HTML、TXT（各最大 10 MB、最大 5 ファイル）',
+ 'reservations.import.acceptedPhotos': '写真（JPG、PNG、WEBP）は AI モデルが読み取ります。',
'reservations.import.parsing': 'ファイルを解析中…',
'reservations.import.previewHeading': '{count} 件の予約が見つかりました',
'reservations.import.previewEmpty': 'アップロードされたファイルから予約を抽出できませんでした。',
diff --git a/shared/src/i18n/ja/settings.ts b/shared/src/i18n/ja/settings.ts
index 647e83c33..5e49bd554 100644
--- a/shared/src/i18n/ja/settings.ts
+++ b/shared/src/i18n/ja/settings.ts
@@ -372,8 +372,8 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'モデルの実行場所 — ローカルのOllamaサーバー、またはOpenAI互換のエンドポイント。',
'settings.aiParsing.apiKey': 'APIキー',
'settings.aiParsing.apiKeyHint': '暗号化して保存されます。現在のキーを保持する場合は空欄のままにしてください。',
- 'settings.aiParsing.multimodal': 'ドキュメントを画像として送信',
- 'settings.aiParsing.multimodalHint': '画像認識対応モデル向け — 抽出したテキストの代わりに元のPDFを送信します。',
+ 'settings.aiParsing.multimodal': 'モデルが画像を読める',
+ 'settings.aiParsing.multimodalHint': '画像を読めるモデルでオンにすると、写真をインポートやスキャンできます。',
'settings.aiParsing.toast.saved': 'AI設定を保存しました',
'settings.aiParsing.toast.saveError': 'AI設定を保存できませんでした',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/ko/admin.ts b/shared/src/i18n/ko/admin.ts
index 165b5b563..c83a1059a 100644
--- a/shared/src/i18n/ko/admin.ts
+++ b/shared/src/i18n/ko/admin.ts
@@ -638,6 +638,11 @@ const admin: TranslationStrings = {
'사용자가 각자 연결한 Dawarich 인스턴스에서 방문 기록과 기록된 경로를 읽어옵니다',
'admin.addons.catalog.llm_parsing.name': 'AI 분석',
'admin.addons.catalog.llm_parsing.description': '기본 분석기가 읽지 못하는 예약을 원하는 AI 모델로 읽어냅니다',
+ 'admin.addons.llm.vision.auto': '자동',
+ 'admin.addons.llm.vision.on': '예',
+ 'admin.addons.llm.vision.off': '아니요',
+ 'admin.addons.llm.vision.hintLocal': '자동은 이 모델이 이미지를 읽을 수 있는지 Ollama 서버에 묻습니다.',
+ 'admin.addons.llm.vision.hintCloud': '클라우드 모델에서 자동은 아니요를 뜻합니다. 이 모델이 이미지를 읽을 수 있으면 예를 선택하세요.',
'admin.addons.enabled': '활성화됨',
'admin.addons.disabled': '비활성화됨',
'admin.addons.type.trip': '여행',
diff --git a/shared/src/i18n/ko/budget.ts b/shared/src/i18n/ko/budget.ts
index f567f64cc..947154036 100644
--- a/shared/src/i18n/ko/budget.ts
+++ b/shared/src/i18n/ko/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': '영수증',
'costs.receiptsTitle': '영수증 및 청구서',
'costs.attachReceipt': '영수증 / 청구서 첨부',
+ 'costs.scan.button': '영수증 스캔',
+ 'costs.scan.reading': '영수증 읽는 중…',
+ 'costs.scan.review': '지출 검토',
+ 'costs.scan.nothingRead': '이 사진에서 영수증을 읽지 못했습니다.',
+ 'costs.scan.failed': '영수증 읽기를 시작하지 못했습니다.',
+ 'costs.scan.title': '영수증 스캔',
+ 'costs.scan.accepted': '영수증 한 장의 사진(JPG, PNG, WEBP, 최대 10MB). AI 모델이 백그라운드에서 읽은 뒤, 저장하기 전에 지출을 확인합니다.',
+ 'costs.scan.dropHere': '영수증 사진을 여기에 놓거나, 클릭해서 선택하거나 촬영하세요',
+ 'costs.scan.dropActive': '놓아서 사진 스캔',
+ 'costs.scan.start': '스캔',
+ 'costs.scan.unsupported': '사진만 스캔할 수 있습니다: JPG, PNG, WEBP.',
'costs.noReceipts': '첨부된 영수증 없음',
'costs.deleteReceipt': '영수증 삭제',
'costs.viewReceipt': '영수증 보기',
diff --git a/shared/src/i18n/ko/reservations.ts b/shared/src/i18n/ko/reservations.ts
index 8a09546e1..a67fcab94 100644
--- a/shared/src/i18n/ko/reservations.ts
+++ b/shared/src/i18n/ko/reservations.ts
@@ -128,6 +128,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': '예약 확인 파일을 여기에 끌어다 놓거나 클릭하여 선택',
'reservations.import.dropActive': '가져올 파일을 여기에 놓으세요',
'reservations.import.acceptedFormats': '허용 형식: EML, PDF, PKPass, HTML, TXT (파일당 최대 10 MB, 최대 5개)',
+ 'reservations.import.acceptedPhotos': '사진(JPG, PNG, WEBP)은 AI 모델이 읽습니다.',
'reservations.import.parsing': '파일 분석 중…',
'reservations.import.previewHeading': '{count}개 예약 발견',
'reservations.import.previewEmpty': '업로드된 파일에서 예약을 추출할 수 없었습니다.',
diff --git a/shared/src/i18n/ko/settings.ts b/shared/src/i18n/ko/settings.ts
index 4148513c9..7beffd94d 100644
--- a/shared/src/i18n/ko/settings.ts
+++ b/shared/src/i18n/ko/settings.ts
@@ -387,8 +387,8 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': '모델이 실행되는 위치 — 로컬 Ollama 서버 또는 OpenAI 호환 엔드포인트입니다.',
'settings.aiParsing.apiKey': 'API 키',
'settings.aiParsing.apiKeyHint': '암호화되어 저장됩니다. 현재 키를 유지하려면 비워 두세요.',
- 'settings.aiParsing.multimodal': '문서를 이미지로 전송',
- 'settings.aiParsing.multimodalHint': '비전 지원 모델용 — 추출된 텍스트 대신 원본 PDF를 전송합니다.',
+ 'settings.aiParsing.multimodal': '모델이 이미지를 읽음',
+ 'settings.aiParsing.multimodalHint': '이미지를 읽는 모델이면 켜세요. 사진을 가져오거나 스캔할 수 있습니다.',
'settings.aiParsing.toast.saved': 'AI 설정이 저장되었습니다',
'settings.aiParsing.toast.saveError': 'AI 설정을 저장할 수 없습니다',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/nl/admin.ts b/shared/src/i18n/nl/admin.ts
index 10a3f8ccd..80aa58c14 100644
--- a/shared/src/i18n/nl/admin.ts
+++ b/shared/src/i18n/nl/admin.ts
@@ -622,6 +622,11 @@ const admin: TranslationStrings = {
'Lees bezoeken en opgenomen routes uit een Dawarich-instantie die elke lezer zelf koppelt',
'admin.addons.catalog.llm_parsing.name': 'AI-analyse',
'admin.addons.catalog.llm_parsing.description': 'Leest boekingen die de ingebouwde parser niet begrijpt, met een AI-model naar keuze',
+ 'admin.addons.llm.vision.auto': 'Automatisch',
+ 'admin.addons.llm.vision.on': 'Ja',
+ 'admin.addons.llm.vision.off': 'Nee',
+ 'admin.addons.llm.vision.hintLocal': 'Automatisch vraagt de Ollama-server of dit model afbeeldingen leest.',
+ 'admin.addons.llm.vision.hintCloud': 'Automatisch betekent nee voor een cloudmodel. Kies Ja als dit model afbeeldingen leest.',
'admin.addons.enabled': 'Ingeschakeld',
'admin.addons.disabled': 'Uitgeschakeld',
'admin.addons.type.trip': 'Reis',
diff --git a/shared/src/i18n/nl/budget.ts b/shared/src/i18n/nl/budget.ts
index f6c19bebf..879e84b2c 100644
--- a/shared/src/i18n/nl/budget.ts
+++ b/shared/src/i18n/nl/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Bonnetjes',
'costs.receiptsTitle': 'Bonnetjes & facturen',
'costs.attachReceipt': 'Bon / factuur bijvoegen',
+ 'costs.scan.button': 'Bon scannen',
+ 'costs.scan.reading': 'Bon wordt gelezen…',
+ 'costs.scan.review': 'Uitgave controleren',
+ 'costs.scan.nothingRead': 'Uit deze foto kon geen bon worden gelezen.',
+ 'costs.scan.failed': 'Het lezen van de bon kon niet worden gestart.',
+ 'costs.scan.title': 'Bon scannen',
+ 'costs.scan.accepted': 'Een foto van één bon (JPG, PNG of WEBP, max. 10 MB). Het AI-model leest hem op de achtergrond; daarna controleer je de uitgave voordat die wordt opgeslagen.',
+ 'costs.scan.dropHere': 'Zet de foto van de bon hier neer, of klik om er een te kiezen of te maken',
+ 'costs.scan.dropActive': 'Laat de foto los om te scannen',
+ 'costs.scan.start': 'Scannen',
+ 'costs.scan.unsupported': 'Alleen een foto kan worden gescand: JPG, PNG of WEBP.',
'costs.noReceipts': 'Geen bonnetjes bijgevoegd',
'costs.deleteReceipt': 'Bonnetje verwijderen',
'costs.viewReceipt': 'Bonnetje bekijken',
diff --git a/shared/src/i18n/nl/reservations.ts b/shared/src/i18n/nl/reservations.ts
index 3703ff469..e472d0949 100644
--- a/shared/src/i18n/nl/reservations.ts
+++ b/shared/src/i18n/nl/reservations.ts
@@ -130,6 +130,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Laat bestanden los om te importeren',
'reservations.import.acceptedFormats':
'Geaccepteerd: EML, PDF, PKPass, HTML, TXT (max. 10 MB per stuk, tot 5 bestanden)',
+ 'reservations.import.acceptedPhotos': "Foto's (JPG, PNG, WEBP) worden door het AI-model gelezen.",
'reservations.import.parsing': 'Bestanden verwerken…',
'reservations.import.previewHeading': '{count} reservering(en) gevonden',
'reservations.import.previewEmpty': 'Er konden geen reserveringen worden geëxtraheerd uit de geüploade bestanden.',
diff --git a/shared/src/i18n/nl/settings.ts b/shared/src/i18n/nl/settings.ts
index 6d976ff4f..b8d901785 100644
--- a/shared/src/i18n/nl/settings.ts
+++ b/shared/src/i18n/nl/settings.ts
@@ -398,9 +398,9 @@ const settings: TranslationStrings = {
'Waar het model draait — een lokale Ollama-server of een OpenAI-compatibel endpoint.',
'settings.aiParsing.apiKey': 'API-sleutel',
'settings.aiParsing.apiKeyHint': 'Versleuteld opgeslagen. Laat leeg om de huidige sleutel te behouden.',
- 'settings.aiParsing.multimodal': 'Documenten als afbeeldingen versturen',
+ 'settings.aiParsing.multimodal': 'Model leest afbeeldingen',
'settings.aiParsing.multimodalHint':
- 'Voor modellen met beeldherkenning — verstuurt de originele PDF in plaats van geëxtraheerde tekst.',
+ 'Zet aan voor een model dat afbeeldingen leest, zodat een foto kan worden geïmporteerd of gescand.',
'settings.aiParsing.toast.saved': 'AI-instellingen opgeslagen',
'settings.aiParsing.toast.saveError': 'AI-instellingen konden niet worden opgeslagen',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/pl/admin.ts b/shared/src/i18n/pl/admin.ts
index 61a9d1075..a26673878 100644
--- a/shared/src/i18n/pl/admin.ts
+++ b/shared/src/i18n/pl/admin.ts
@@ -621,6 +621,11 @@ const admin: TranslationStrings = {
'Odczytuj pobyty i zapisane trasy z instancji Dawarich, którą każdy użytkownik podłącza sam',
'admin.addons.catalog.llm_parsing.name': 'Analiza AI',
'admin.addons.catalog.llm_parsing.description': 'Odczytuje rezerwacje, których nie rozumie wbudowany parser, przy użyciu wybranego modelu AI',
+ 'admin.addons.llm.vision.auto': 'Automatycznie',
+ 'admin.addons.llm.vision.on': 'Tak',
+ 'admin.addons.llm.vision.off': 'Nie',
+ 'admin.addons.llm.vision.hintLocal': 'Automatycznie pyta serwer Ollama, czy ten model czyta obrazy.',
+ 'admin.addons.llm.vision.hintCloud': 'Automatycznie oznacza „nie” dla modelu w chmurze. Wybierz Tak, jeśli ten model czyta obrazy.',
'admin.addons.enabled': 'Włączone',
'admin.addons.disabled': 'Wyłączone',
'admin.addons.type.trip': 'Podróż',
diff --git a/shared/src/i18n/pl/budget.ts b/shared/src/i18n/pl/budget.ts
index 26438eaef..68663cec7 100644
--- a/shared/src/i18n/pl/budget.ts
+++ b/shared/src/i18n/pl/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Paragony',
'costs.receiptsTitle': 'Paragony i faktury',
'costs.attachReceipt': 'Załącz paragon / fakturę',
+ 'costs.scan.button': 'Zeskanuj paragon',
+ 'costs.scan.reading': 'Odczytywanie paragonu…',
+ 'costs.scan.review': 'Sprawdź wydatek',
+ 'costs.scan.nothingRead': 'Z tego zdjęcia nie udało się odczytać paragonu.',
+ 'costs.scan.failed': 'Nie udało się rozpocząć odczytu paragonu.',
+ 'costs.scan.title': 'Zeskanuj paragon',
+ 'costs.scan.accepted': 'Zdjęcie jednego paragonu (JPG, PNG lub WEBP, do 10 MB). Model AI odczyta je w tle, a potem sprawdzisz wydatek przed zapisaniem.',
+ 'costs.scan.dropHere': 'Upuść tu zdjęcie paragonu albo kliknij, aby je wybrać lub zrobić',
+ 'costs.scan.dropActive': 'Upuść zdjęcie, aby je zeskanować',
+ 'costs.scan.start': 'Skanuj',
+ 'costs.scan.unsupported': 'Skanować można tylko zdjęcie: JPG, PNG lub WEBP.',
'costs.noReceipts': 'Brak załączonych paragonów',
'costs.deleteReceipt': 'Usuń paragon',
'costs.viewReceipt': 'Zobacz paragon',
diff --git a/shared/src/i18n/pl/reservations.ts b/shared/src/i18n/pl/reservations.ts
index cec29211b..02145d5ee 100644
--- a/shared/src/i18n/pl/reservations.ts
+++ b/shared/src/i18n/pl/reservations.ts
@@ -129,6 +129,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Upuść pliki potwierdzeń rezerwacji tutaj lub kliknij, aby wybrać',
'reservations.import.dropActive': 'Upuść pliki, aby zaimportować',
'reservations.import.acceptedFormats': 'Akceptowane: EML, PDF, PKPass, HTML, TXT (maks. 10 MB każdy, do 5 plików)',
+ 'reservations.import.acceptedPhotos': 'Zdjęcia (JPG, PNG, WEBP) odczytuje model AI.',
'reservations.import.parsing': 'Przetwarzanie plików…',
'reservations.import.previewHeading': 'Znaleziono {count} rezerwację/rezerwacje',
'reservations.import.previewEmpty': 'Nie udało się wyodrębnić rezerwacji z przesłanych plików.',
diff --git a/shared/src/i18n/pl/settings.ts b/shared/src/i18n/pl/settings.ts
index 71ed56522..98f80d339 100644
--- a/shared/src/i18n/pl/settings.ts
+++ b/shared/src/i18n/pl/settings.ts
@@ -397,9 +397,9 @@ const settings: TranslationStrings = {
'Miejsce, w którym działa model — lokalny serwer Ollama lub punkt końcowy zgodny z OpenAI.',
'settings.aiParsing.apiKey': 'Klucz API',
'settings.aiParsing.apiKeyHint': 'Przechowywany w postaci zaszyfrowanej. Pozostaw puste, aby zachować bieżący klucz.',
- 'settings.aiParsing.multimodal': 'Wysyłaj dokumenty jako obrazy',
+ 'settings.aiParsing.multimodal': 'Model czyta obrazy',
'settings.aiParsing.multimodalHint':
- 'Dla modeli obsługujących obraz — wysyła oryginalny plik PDF zamiast wyodrębnionego tekstu.',
+ 'Włącz dla modelu, który czyta obrazy, aby można było importować lub skanować zdjęcia.',
'settings.aiParsing.toast.saved': 'Zapisano ustawienia AI',
'settings.aiParsing.toast.saveError': 'Nie udało się zapisać ustawień AI',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/ru/admin.ts b/shared/src/i18n/ru/admin.ts
index ab912c4d3..da290b4df 100644
--- a/shared/src/i18n/ru/admin.ts
+++ b/shared/src/i18n/ru/admin.ts
@@ -626,6 +626,11 @@ const admin: TranslationStrings = {
'Чтение посещений и записанных маршрутов из экземпляра Dawarich, который каждый пользователь подключает сам',
'admin.addons.catalog.llm_parsing.name': 'Разбор с помощью ИИ',
'admin.addons.catalog.llm_parsing.description': 'Читает брони, которые не понимает встроенный анализатор, с помощью выбранной вами модели ИИ',
+ 'admin.addons.llm.vision.auto': 'Автоматически',
+ 'admin.addons.llm.vision.on': 'Да',
+ 'admin.addons.llm.vision.off': 'Нет',
+ 'admin.addons.llm.vision.hintLocal': '«Автоматически» спрашивает сервер Ollama, читает ли эта модель изображения.',
+ 'admin.addons.llm.vision.hintCloud': 'Для облачной модели «Автоматически» означает «нет». Выберите «Да», если эта модель читает изображения.',
'admin.addons.enabled': 'Включено',
'admin.addons.disabled': 'Отключено',
'admin.addons.type.trip': 'Поездка',
diff --git a/shared/src/i18n/ru/budget.ts b/shared/src/i18n/ru/budget.ts
index 0ddc619af..344fc96b1 100644
--- a/shared/src/i18n/ru/budget.ts
+++ b/shared/src/i18n/ru/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Чеки',
'costs.receiptsTitle': 'Чеки и счета',
'costs.attachReceipt': 'Прикрепить чек / счет',
+ 'costs.scan.button': 'Сканировать чек',
+ 'costs.scan.reading': 'Чтение чека…',
+ 'costs.scan.review': 'Проверить расход',
+ 'costs.scan.nothingRead': 'Не удалось прочитать чек с этой фотографии.',
+ 'costs.scan.failed': 'Не удалось начать чтение чека.',
+ 'costs.scan.title': 'Сканировать чек',
+ 'costs.scan.accepted': 'Фото одного чека (JPG, PNG или WEBP, до 10 МБ). Модель ИИ прочитает его в фоне, а затем вы проверите расход перед сохранением.',
+ 'costs.scan.dropHere': 'Перетащите сюда фото чека или нажмите, чтобы выбрать или сделать снимок',
+ 'costs.scan.dropActive': 'Отпустите фото, чтобы отсканировать его',
+ 'costs.scan.start': 'Сканировать',
+ 'costs.scan.unsupported': 'Сканировать можно только фото: JPG, PNG или WEBP.',
'costs.noReceipts': 'Нет прикрепленных чеков',
'costs.deleteReceipt': 'Удалить чек',
'costs.viewReceipt': 'Просмотреть чек',
diff --git a/shared/src/i18n/ru/reservations.ts b/shared/src/i18n/ru/reservations.ts
index 4b2292d56..61e7796c0 100644
--- a/shared/src/i18n/ru/reservations.ts
+++ b/shared/src/i18n/ru/reservations.ts
@@ -128,6 +128,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Перетащите файлы подтверждений бронирования сюда или нажмите для выбора',
'reservations.import.dropActive': 'Отпустите файлы для импорта',
'reservations.import.acceptedFormats': 'Принимаются: EML, PDF, PKPass, HTML, TXT (макс. 10 МБ каждый, до 5 файлов)',
+ 'reservations.import.acceptedPhotos': 'Фотографии (JPG, PNG, WEBP) читает модель ИИ.',
'reservations.import.parsing': 'Обработка файлов…',
'reservations.import.previewHeading': 'Найдено {count} бронирование(й)',
'reservations.import.previewEmpty': 'Из загруженных файлов не удалось извлечь бронирования.',
diff --git a/shared/src/i18n/ru/settings.ts b/shared/src/i18n/ru/settings.ts
index 0bea98457..6de2fb1d7 100644
--- a/shared/src/i18n/ru/settings.ts
+++ b/shared/src/i18n/ru/settings.ts
@@ -398,9 +398,9 @@ const settings: TranslationStrings = {
'Где работает модель — локальный сервер Ollama или конечная точка, совместимая с OpenAI.',
'settings.aiParsing.apiKey': 'Ключ API',
'settings.aiParsing.apiKeyHint': 'Хранится в зашифрованном виде. Оставьте поле пустым, чтобы сохранить текущий ключ.',
- 'settings.aiParsing.multimodal': 'Отправлять документы как изображения',
+ 'settings.aiParsing.multimodal': 'Модель читает изображения',
'settings.aiParsing.multimodalHint':
- 'Для моделей с поддержкой изображений — отправляет исходный PDF вместо извлечённого текста.',
+ 'Включите для модели, которая читает изображения, чтобы фотографию можно было импортировать или отсканировать.',
'settings.aiParsing.toast.saved': 'Настройки ИИ сохранены',
'settings.aiParsing.toast.saveError': 'Не удалось сохранить настройки ИИ',
'settings.tabs.appearance': 'Внешний вид',
diff --git a/shared/src/i18n/sk/admin.ts b/shared/src/i18n/sk/admin.ts
index 4d7d182ec..81c0c899f 100644
--- a/shared/src/i18n/sk/admin.ts
+++ b/shared/src/i18n/sk/admin.ts
@@ -676,6 +676,11 @@ const admin: TranslationStrings = {
'admin.addons.catalog.llm_parsing.name': 'Analýza pomocou AI',
'admin.addons.catalog.llm_parsing.description':
'Prečíta rezervácie, ktoré vstavaný analyzátor nezvládne, pomocou zvoleného modelu AI',
+ 'admin.addons.llm.vision.auto': 'Automaticky',
+ 'admin.addons.llm.vision.on': 'Áno',
+ 'admin.addons.llm.vision.off': 'Nie',
+ 'admin.addons.llm.vision.hintLocal': 'Automaticky sa opýta servera Ollama, či tento model číta obrázky.',
+ 'admin.addons.llm.vision.hintCloud': 'Pri cloudovom modeli znamená Automaticky nie. Zvoľte Áno, ak tento model číta obrázky.',
'admin.addons.enabled': 'Povolené',
'admin.addons.disabled': 'Zakázané',
'admin.addons.type.trip': 'Cesta',
diff --git a/shared/src/i18n/sk/budget.ts b/shared/src/i18n/sk/budget.ts
index daa83260b..c81a44b7b 100644
--- a/shared/src/i18n/sk/budget.ts
+++ b/shared/src/i18n/sk/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Účtenky',
'costs.receiptsTitle': 'Účtenky a faktúry',
'costs.attachReceipt': 'Priložiť účtenku / faktúru',
+ 'costs.scan.button': 'Naskenovať účtenku',
+ 'costs.scan.reading': 'Číta sa účtenka…',
+ 'costs.scan.review': 'Skontrolovať výdavok',
+ 'costs.scan.nothingRead': 'Z tejto fotografie sa nepodarilo prečítať účtenku.',
+ 'costs.scan.failed': 'Čítanie účtenky sa nepodarilo spustiť.',
+ 'costs.scan.title': 'Naskenovať účtenku',
+ 'costs.scan.accepted': 'Fotografia jednej účtenky (JPG, PNG alebo WEBP, do 10 MB). Model AI ju prečíta na pozadí a vy potom výdavok pred uložením skontrolujete.',
+ 'costs.scan.dropHere': 'Presuňte sem fotografiu účtenky alebo kliknite a vyberte či odfoťte novú',
+ 'costs.scan.dropActive': 'Pustite fotografiu na naskenovanie',
+ 'costs.scan.start': 'Skenovať',
+ 'costs.scan.unsupported': 'Skenovať sa dá len fotografia: JPG, PNG alebo WEBP.',
'costs.noReceipts': 'Žiadne priložené účtenky',
'costs.deleteReceipt': 'Odobrať účtenku',
'costs.viewReceipt': 'Zobraziť účtenku',
diff --git a/shared/src/i18n/sk/reservations.ts b/shared/src/i18n/sk/reservations.ts
index bd000dd3a..b7c786040 100644
--- a/shared/src/i18n/sk/reservations.ts
+++ b/shared/src/i18n/sk/reservations.ts
@@ -134,6 +134,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Presuňte sem súbory s potvrdením rezervácie alebo kliknite pre výber',
'reservations.import.dropActive': 'Pustite súbory pre import',
'reservations.import.acceptedFormats': 'Prijímané: EML, PDF, PKPass, HTML, TXT (max. 10 MB každý, až 5 súborov)',
+ 'reservations.import.acceptedPhotos': 'Fotografie (JPG, PNG, WEBP) číta model AI.',
'reservations.import.parsing': 'Spracúvajú sa súbory…',
'reservations.import.previewHeading': 'Nájdených {count} rezervácií',
'reservations.import.previewEmpty': 'Z nahraných súborov sa nepodarilo extrahovať žiadne rezervácie.',
diff --git a/shared/src/i18n/sk/settings.ts b/shared/src/i18n/sk/settings.ts
index d5c572f3a..1e1a4a2f9 100644
--- a/shared/src/i18n/sk/settings.ts
+++ b/shared/src/i18n/sk/settings.ts
@@ -404,9 +404,9 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'Kde model beží — lokálny server Ollama alebo koncový bod kompatibilný s OpenAI.',
'settings.aiParsing.apiKey': 'API kľúč',
'settings.aiParsing.apiKeyHint': 'Ukladá sa šifrovane. Ponechajte prázdne pre zachovanie aktuálneho kľúča.',
- 'settings.aiParsing.multimodal': 'Odosielať dokumenty ako obrázky',
+ 'settings.aiParsing.multimodal': 'Model číta obrázky',
'settings.aiParsing.multimodalHint':
- 'Pre modely so schopnosťou spracovať obraz — odošle pôvodné PDF namiesto extrahovaného textu.',
+ 'Zapnite pre model, ktorý číta obrázky, aby sa fotografia dala importovať alebo naskenovať.',
'settings.aiParsing.toast.saved': 'Nastavenia AI uložené',
'settings.aiParsing.toast.saveError': 'Nastavenia AI sa nepodarilo uložiť',
'settings.tabs.appearance': 'Vzhľad',
diff --git a/shared/src/i18n/sv/admin.ts b/shared/src/i18n/sv/admin.ts
index 9cec67d2c..f09058ed4 100644
--- a/shared/src/i18n/sv/admin.ts
+++ b/shared/src/i18n/sv/admin.ts
@@ -687,6 +687,11 @@ const admin: TranslationStrings = {
'Läs vistelser och inspelade rutter från en Dawarich-instans som varje läsare ansluter själv',
'admin.addons.catalog.llm_parsing.name': 'AI-tolkning',
'admin.addons.catalog.llm_parsing.description': 'Läser bokningar som den inbyggda tolken inte klarar, med en AI-modell du väljer',
+ 'admin.addons.llm.vision.auto': 'Automatiskt',
+ 'admin.addons.llm.vision.on': 'Ja',
+ 'admin.addons.llm.vision.off': 'Nej',
+ 'admin.addons.llm.vision.hintLocal': 'Automatiskt frågar Ollama-servern om modellen läser bilder.',
+ 'admin.addons.llm.vision.hintCloud': 'Automatiskt betyder nej för en molnmodell. Välj Ja om modellen läser bilder.',
'admin.addons.enabled': 'Aktiverad',
'admin.addons.disabled': 'Inaktiverad',
'admin.addons.type.trip': 'Resa',
diff --git a/shared/src/i18n/sv/budget.ts b/shared/src/i18n/sv/budget.ts
index ae26db6b0..065ae0a84 100644
--- a/shared/src/i18n/sv/budget.ts
+++ b/shared/src/i18n/sv/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Kvitton',
'costs.receiptsTitle': 'Kvitton & fakturor',
'costs.attachReceipt': 'Bifoga kvitto / faktura',
+ 'costs.scan.button': 'Skanna kvitto',
+ 'costs.scan.reading': 'Läser kvittot…',
+ 'costs.scan.review': 'Granska utgiften',
+ 'costs.scan.nothingRead': 'Inget kvitto kunde läsas från fotot.',
+ 'costs.scan.failed': 'Det gick inte att börja läsa kvittot.',
+ 'costs.scan.title': 'Skanna kvitto',
+ 'costs.scan.accepted': 'Ett foto av ett kvitto (JPG, PNG eller WEBP, högst 10 MB). AI-modellen läser det i bakgrunden, sedan granskar du utgiften innan den sparas.',
+ 'costs.scan.dropHere': 'Släpp kvittofotot här, eller klicka för att välja eller ta ett',
+ 'costs.scan.dropActive': 'Släpp fotot för att skanna det',
+ 'costs.scan.start': 'Skanna',
+ 'costs.scan.unsupported': 'Bara ett foto kan skannas: JPG, PNG eller WEBP.',
'costs.noReceipts': 'Inga bifogade kvitton',
'costs.deleteReceipt': 'Ta bort kvitto',
'costs.viewReceipt': 'Visa kvitto',
diff --git a/shared/src/i18n/sv/reservations.ts b/shared/src/i18n/sv/reservations.ts
index da7bd5ad3..bb1d2a958 100644
--- a/shared/src/i18n/sv/reservations.ts
+++ b/shared/src/i18n/sv/reservations.ts
@@ -130,6 +130,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Dra och släpp filerna som ska importeras',
'reservations.import.acceptedFormats':
'Godkända filformat: EML, PDF, PKPass, HTML, TXT (högst 10 MB per fil, upp till 5 filer)',
+ 'reservations.import.acceptedPhotos': 'Foton (JPG, PNG, WEBP) läses av AI-modellen.',
'reservations.import.parsing': 'Analyserar filer…',
'reservations.import.previewHeading': '{count} reservation(er) hittades',
'reservations.import.previewEmpty': 'Inga bokningar kunde extraheras från de uppladdade filerna.',
diff --git a/shared/src/i18n/sv/settings.ts b/shared/src/i18n/sv/settings.ts
index 8d2a15dcb..26c297551 100644
--- a/shared/src/i18n/sv/settings.ts
+++ b/shared/src/i18n/sv/settings.ts
@@ -396,9 +396,9 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'Var modellen körs — en lokal Ollama-server eller en OpenAI-kompatibel slutpunkt.',
'settings.aiParsing.apiKey': 'API-nyckel',
'settings.aiParsing.apiKeyHint': 'Lagras krypterad. Lämna tomt för att behålla den nuvarande nyckeln.',
- 'settings.aiParsing.multimodal': 'Skicka dokument som bilder',
+ 'settings.aiParsing.multimodal': 'Modellen läser bilder',
'settings.aiParsing.multimodalHint':
- 'För modeller med bildstöd — skickar original-PDF:en istället för extraherad text.',
+ 'Slå på för en modell som läser bilder, så att ett foto kan importeras eller skannas.',
'settings.aiParsing.toast.saved': 'AI-inställningar sparade',
'settings.aiParsing.toast.saveError': 'Kunde inte spara AI-inställningarna',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/th/admin.ts b/shared/src/i18n/th/admin.ts
index 3a350655c..b3fb6d6a4 100644
--- a/shared/src/i18n/th/admin.ts
+++ b/shared/src/i18n/th/admin.ts
@@ -598,6 +598,11 @@ const admin: TranslationStrings = {
'admin.addons.catalog.airtrail.description': "ซิงค์เที่ยวบินจากอินสแตนซ์ AirTrail ของคุณ",
'admin.addons.catalog.llm_parsing.name': "การแยกวิเคราะห์ AI",
'admin.addons.catalog.llm_parsing.description': "อ่านการจองที่ parser ในตัวไม่สามารถทำได้ โดยใช้โมเดล AI ที่คุณเลือก",
+ 'admin.addons.llm.vision.auto': "อัตโนมัติ",
+ 'admin.addons.llm.vision.on': "ใช่",
+ 'admin.addons.llm.vision.off': "ไม่",
+ 'admin.addons.llm.vision.hintLocal': "โหมดอัตโนมัติจะถามเซิร์ฟเวอร์ Ollama ว่าโมเดลนี้อ่านรูปภาพได้หรือไม่",
+ 'admin.addons.llm.vision.hintCloud': "สำหรับโมเดลบนคลาวด์ โหมดอัตโนมัติหมายถึงไม่ เลือกใช่หากโมเดลนี้อ่านรูปภาพได้",
'admin.addons.enabled': "เปิดใช้งานแล้ว",
'admin.addons.disabled': "ปิดการใช้งาน",
'admin.addons.type.trip': "การเดินทาง",
diff --git a/shared/src/i18n/th/budget.ts b/shared/src/i18n/th/budget.ts
index 058467625..fba6ff704 100644
--- a/shared/src/i18n/th/budget.ts
+++ b/shared/src/i18n/th/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': "ใบเสร็จ",
'costs.receiptsTitle': "ใบเสร็จและใบแจ้งหนี้",
'costs.attachReceipt': "แนบใบเสร็จ / ใบแจ้งหนี้",
+ 'costs.scan.button': "สแกนใบเสร็จ",
+ 'costs.scan.reading': "กำลังอ่านใบเสร็จ…",
+ 'costs.scan.review': "ตรวจสอบค่าใช้จ่าย",
+ 'costs.scan.nothingRead': "อ่านใบเสร็จจากรูปนี้ไม่ได้",
+ 'costs.scan.failed': "ไม่สามารถเริ่มอ่านใบเสร็จได้",
+ 'costs.scan.title': "สแกนใบเสร็จ",
+ 'costs.scan.accepted': "รูปถ่ายใบเสร็จหนึ่งใบ (JPG, PNG หรือ WEBP ไม่เกิน 10 MB) โมเดล AI จะอ่านอยู่เบื้องหลัง จากนั้นคุณตรวจสอบค่าใช้จ่ายก่อนบันทึก",
+ 'costs.scan.dropHere': "วางรูปใบเสร็จที่นี่ หรือคลิกเพื่อเลือกหรือถ่ายรูป",
+ 'costs.scan.dropActive': "ปล่อยรูปเพื่อสแกน",
+ 'costs.scan.start': "สแกน",
+ 'costs.scan.unsupported': "สแกนได้เฉพาะรูปถ่าย: JPG, PNG หรือ WEBP",
'costs.noReceipts': "ยังไม่มีใบเสร็จแนบ",
'costs.deleteReceipt': "ลบใบเสร็จ",
'costs.viewReceipt': "ดูใบเสร็จ",
diff --git a/shared/src/i18n/th/reservations.ts b/shared/src/i18n/th/reservations.ts
index 5febfde82..2f3f3168c 100644
--- a/shared/src/i18n/th/reservations.ts
+++ b/shared/src/i18n/th/reservations.ts
@@ -133,6 +133,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': "วางไฟล์ยืนยันการจองที่นี่ หรือคลิกเพื่อเลือก",
'reservations.import.dropActive': "วางไฟล์ที่จะนำเข้า",
'reservations.import.acceptedFormats': "ยอมรับ: EML, PDF, PKPass, HTML, TXT (สูงสุด 10 MB ต่อไฟล์ สูงสุด 5 ไฟล์)",
+ 'reservations.import.acceptedPhotos': "โมเดล AI จะอ่านรูปถ่าย (JPG, PNG, WEBP)",
'reservations.import.parsing': "กำลังแยกวิเคราะห์ไฟล์...",
'reservations.import.previewHeading': "พบการจอง {count} แล้ว",
'reservations.import.previewEmpty': "ไม่สามารถแยกการจองออกจากไฟล์ที่อัปโหลดได้",
diff --git a/shared/src/i18n/th/settings.ts b/shared/src/i18n/th/settings.ts
index 48396d6c6..3cf0e112d 100644
--- a/shared/src/i18n/th/settings.ts
+++ b/shared/src/i18n/th/settings.ts
@@ -391,8 +391,8 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': "ตำแหน่งที่โมเดลทำงาน — เซิร์ฟเวอร์ Ollama ในเครื่องหรือจุดสิ้นสุดที่เข้ากันได้กับ OpenAI",
'settings.aiParsing.apiKey': "คีย์เอพีไอ",
'settings.aiParsing.apiKeyHint': "จัดเก็บแบบเข้ารหัส เว้นว่างไว้เพื่อเก็บคีย์ปัจจุบันไว้",
- 'settings.aiParsing.multimodal': "ส่งเอกสารเป็นรูปภาพ",
- 'settings.aiParsing.multimodalHint': "สำหรับรุ่นที่มองเห็นได้ — ส่ง PDF ต้นฉบับแทนข้อความที่แยกออกมา",
+ 'settings.aiParsing.multimodal': "โมเดลอ่านรูปภาพได้",
+ 'settings.aiParsing.multimodalHint': "เปิดสำหรับโมเดลที่อ่านรูปภาพได้ เพื่อให้นำเข้าหรือสแกนรูปถ่ายได้",
'settings.aiParsing.toast.saved': "บันทึกการตั้งค่า AI แล้ว",
'settings.aiParsing.toast.saveError': "ไม่สามารถบันทึกการตั้งค่า AI",
'settings.tabs.appearance': "ลักษณะที่ปรากฏ",
diff --git a/shared/src/i18n/tr/admin.ts b/shared/src/i18n/tr/admin.ts
index ff7a690b6..4db023778 100644
--- a/shared/src/i18n/tr/admin.ts
+++ b/shared/src/i18n/tr/admin.ts
@@ -672,6 +672,11 @@ const admin: TranslationStrings = {
'Her kullanıcının kendi bağladığı bir Dawarich sunucusundan ziyaretleri ve kaydedilen rotaları okuyun',
'admin.addons.catalog.llm_parsing.name': 'Yapay zekâ ile ayrıştırma',
'admin.addons.catalog.llm_parsing.description': 'Yerleşik ayrıştırıcının okuyamadığı rezervasyonları, seçtiğiniz yapay zekâ modeliyle okur',
+ 'admin.addons.llm.vision.auto': 'Otomatik',
+ 'admin.addons.llm.vision.on': 'Evet',
+ 'admin.addons.llm.vision.off': 'Hayır',
+ 'admin.addons.llm.vision.hintLocal': 'Otomatik, bu modelin görselleri okuyup okumadığını Ollama sunucusuna sorar.',
+ 'admin.addons.llm.vision.hintCloud': 'Bulut modeli için Otomatik, hayır anlamına gelir. Bu model görselleri okuyorsa Evet’i seç.',
'admin.addons.enabled': 'Etkinleştirilmiş',
'admin.addons.disabled': 'Engelli',
'admin.addons.type.trip': 'Seyahat',
diff --git a/shared/src/i18n/tr/budget.ts b/shared/src/i18n/tr/budget.ts
index 7de4cbeab..dd3ec9d97 100644
--- a/shared/src/i18n/tr/budget.ts
+++ b/shared/src/i18n/tr/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Fişler',
'costs.receiptsTitle': 'Fişler ve Faturalar',
'costs.attachReceipt': 'Fiş / fatura ekle',
+ 'costs.scan.button': 'Fiş tara',
+ 'costs.scan.reading': 'Fiş okunuyor…',
+ 'costs.scan.review': 'Harcamayı gözden geçir',
+ 'costs.scan.nothingRead': 'Bu fotoğraftan fiş okunamadı.',
+ 'costs.scan.failed': 'Fiş okuma başlatılamadı.',
+ 'costs.scan.title': 'Fiş tara',
+ 'costs.scan.accepted': 'Tek bir fişin fotoğrafı (JPG, PNG veya WEBP, en fazla 10 MB). Yapay zekâ modeli onu arka planda okur, sonra harcamayı kaydetmeden önce kontrol edersin.',
+ 'costs.scan.dropHere': 'Fiş fotoğrafını buraya bırak ya da seçmek veya çekmek için tıkla',
+ 'costs.scan.dropActive': 'Taramak için fotoğrafı bırak',
+ 'costs.scan.start': 'Tara',
+ 'costs.scan.unsupported': 'Yalnızca fotoğraf taranabilir: JPG, PNG veya WEBP.',
'costs.noReceipts': 'Ekli fiş yok',
'costs.deleteReceipt': 'Fişi kaldır',
'costs.viewReceipt': 'Fişi görüntüle',
diff --git a/shared/src/i18n/tr/reservations.ts b/shared/src/i18n/tr/reservations.ts
index 2a75016d5..8d0466b9d 100644
--- a/shared/src/i18n/tr/reservations.ts
+++ b/shared/src/i18n/tr/reservations.ts
@@ -130,6 +130,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'İçe aktarmak için dosyaları bırakın',
'reservations.import.acceptedFormats':
'Kabul edilenler: EML, PDF, PKPass, HTML, TXT (her biri maks. 10 MB, en fazla 5 dosya)',
+ 'reservations.import.acceptedPhotos': 'Fotoğrafları (JPG, PNG, WEBP) yapay zekâ modeli okur.',
'reservations.import.parsing': 'Dosyalar işleniyor…',
'reservations.import.previewHeading': '{count} rezervasyon bulundu',
'reservations.import.previewEmpty': 'Yüklenen dosyalardan hiçbir rezervasyon çıkarılamadı.',
diff --git a/shared/src/i18n/tr/settings.ts b/shared/src/i18n/tr/settings.ts
index 91be392d9..efea109d4 100644
--- a/shared/src/i18n/tr/settings.ts
+++ b/shared/src/i18n/tr/settings.ts
@@ -395,9 +395,9 @@ const settings: TranslationStrings = {
'Modelin çalıştığı yer — yerel bir Ollama sunucusu veya OpenAI uyumlu bir uç nokta.',
'settings.aiParsing.apiKey': 'API anahtarı',
'settings.aiParsing.apiKeyHint': 'Şifrelenmiş olarak saklanır. Mevcut anahtarı korumak için boş bırak.',
- 'settings.aiParsing.multimodal': 'Belgeleri görsel olarak gönder',
+ 'settings.aiParsing.multimodal': 'Model görselleri okur',
'settings.aiParsing.multimodalHint':
- "Görüntü işleyebilen modeller için — çıkarılan metin yerine orijinal PDF'yi gönderir.",
+ 'Görselleri okuyan bir model için aç; böylece bir fotoğraf içe aktarılabilir veya taranabilir.',
'settings.aiParsing.toast.saved': 'Yapay zekâ ayarları kaydedildi',
'settings.aiParsing.toast.saveError': 'Yapay zekâ ayarları kaydedilemedi',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/uk/admin.ts b/shared/src/i18n/uk/admin.ts
index 9498c9aac..74743b68e 100644
--- a/shared/src/i18n/uk/admin.ts
+++ b/shared/src/i18n/uk/admin.ts
@@ -623,6 +623,11 @@ const admin: TranslationStrings = {
'Читає перебування та записані маршрути із сервера Dawarich, який кожен користувач підключає сам',
'admin.addons.catalog.llm_parsing.name': 'Розбір за допомогою ШІ',
'admin.addons.catalog.llm_parsing.description': 'Читає бронювання, які не розпізнає вбудований аналізатор, за допомогою обраної вами моделі ШІ',
+ 'admin.addons.llm.vision.auto': 'Автоматично',
+ 'admin.addons.llm.vision.on': 'Так',
+ 'admin.addons.llm.vision.off': 'Ні',
+ 'admin.addons.llm.vision.hintLocal': '«Автоматично» запитує сервер Ollama, чи читає ця модель зображення.',
+ 'admin.addons.llm.vision.hintCloud': 'Для хмарної моделі «Автоматично» означає «ні». Виберіть «Так», якщо ця модель читає зображення.',
'admin.addons.enabled': 'Увімкнено',
'admin.addons.disabled': 'Вимкнено',
'admin.addons.type.trip': 'Поїздка',
diff --git a/shared/src/i18n/uk/budget.ts b/shared/src/i18n/uk/budget.ts
index 10a3887ce..7c1ed5303 100644
--- a/shared/src/i18n/uk/budget.ts
+++ b/shared/src/i18n/uk/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Чеки',
'costs.receiptsTitle': 'Чеки та рахунки',
'costs.attachReceipt': 'Додати чек / рахунок',
+ 'costs.scan.button': 'Сканувати чек',
+ 'costs.scan.reading': 'Читання чека…',
+ 'costs.scan.review': 'Перевірити витрату',
+ 'costs.scan.nothingRead': 'Не вдалося прочитати чек із цієї фотографії.',
+ 'costs.scan.failed': 'Не вдалося почати читання чека.',
+ 'costs.scan.title': 'Сканувати чек',
+ 'costs.scan.accepted': 'Фото одного чека (JPG, PNG або WEBP, до 10 МБ). Модель ШІ прочитає його у фоні, а потім ви перевірите витрату перед збереженням.',
+ 'costs.scan.dropHere': 'Перетягніть сюди фото чека або натисніть, щоб вибрати чи зробити знімок',
+ 'costs.scan.dropActive': 'Відпустіть фото, щоб відсканувати його',
+ 'costs.scan.start': 'Сканувати',
+ 'costs.scan.unsupported': 'Сканувати можна лише фото: JPG, PNG або WEBP.',
'costs.noReceipts': 'Немає доданих чеків',
'costs.deleteReceipt': 'Видалити чек',
'costs.viewReceipt': 'Переглянути чек',
diff --git a/shared/src/i18n/uk/reservations.ts b/shared/src/i18n/uk/reservations.ts
index 9c85e856f..6dccff6b5 100644
--- a/shared/src/i18n/uk/reservations.ts
+++ b/shared/src/i18n/uk/reservations.ts
@@ -128,6 +128,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': 'Перетягніть файли підтверджень бронювання сюди або натисніть для вибору',
'reservations.import.dropActive': 'Відпустіть файли для імпорту',
'reservations.import.acceptedFormats': 'Підтримуються: EML, PDF, PKPass, HTML, TXT (макс. 10 МБ кожен, до 5 файлів)',
+ 'reservations.import.acceptedPhotos': 'Фотографії (JPG, PNG, WEBP) читає модель ШІ.',
'reservations.import.parsing': 'Обробка файлів…',
'reservations.import.previewHeading': 'Знайдено {count} бронювання(нь)',
'reservations.import.previewEmpty': 'З завантажених файлів не вдалося витягти бронювання.',
diff --git a/shared/src/i18n/uk/settings.ts b/shared/src/i18n/uk/settings.ts
index 9bf1b779a..9a98c99c0 100644
--- a/shared/src/i18n/uk/settings.ts
+++ b/shared/src/i18n/uk/settings.ts
@@ -398,9 +398,9 @@ const settings: TranslationStrings = {
'settings.aiParsing.apiKey': 'Ключ API',
'settings.aiParsing.apiKeyHint':
'Зберігається в зашифрованому вигляді. Залиште порожнім, щоб зберегти поточний ключ.',
- 'settings.aiParsing.multimodal': 'Надсилати документи як зображення',
+ 'settings.aiParsing.multimodal': 'Модель читає зображення',
'settings.aiParsing.multimodalHint':
- 'Для моделей із підтримкою зображень — надсилає оригінальний PDF замість вилученого тексту.',
+ 'Увімкніть для моделі, яка читає зображення, щоб фотографію можна було імпортувати чи відсканувати.',
'settings.aiParsing.toast.saved': 'Налаштування ШІ збережено',
'settings.aiParsing.toast.saveError': 'Не вдалося зберегти налаштування ШІ',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/vi/admin.ts b/shared/src/i18n/vi/admin.ts
index 703175936..3cdbb92f6 100644
--- a/shared/src/i18n/vi/admin.ts
+++ b/shared/src/i18n/vi/admin.ts
@@ -628,6 +628,11 @@ const admin: TranslationStrings = {
'Đọc các điểm lưu trú và lộ trình đã ghi từ phiên bản Dawarich mà mỗi người tự kết nối',
'admin.addons.catalog.llm_parsing.name': 'Phân tích bằng AI',
'admin.addons.catalog.llm_parsing.description': 'Đọc các đặt chỗ mà bộ phân tích tích hợp không hiểu, bằng mô hình AI bạn chọn',
+ 'admin.addons.llm.vision.auto': 'Tự động',
+ 'admin.addons.llm.vision.on': 'Có',
+ 'admin.addons.llm.vision.off': 'Không',
+ 'admin.addons.llm.vision.hintLocal': 'Tự động sẽ hỏi máy chủ Ollama xem mô hình này có đọc được hình ảnh không.',
+ 'admin.addons.llm.vision.hintCloud': 'Với mô hình đám mây, Tự động nghĩa là không. Chọn Có nếu mô hình này đọc được hình ảnh.',
'admin.addons.enabled': 'Đã bật',
'admin.addons.disabled': 'Tắt',
'admin.addons.type.trip': 'Chuyến đi',
diff --git a/shared/src/i18n/vi/budget.ts b/shared/src/i18n/vi/budget.ts
index 6a2eff7e9..62f875b44 100644
--- a/shared/src/i18n/vi/budget.ts
+++ b/shared/src/i18n/vi/budget.ts
@@ -154,6 +154,17 @@ const budget: TranslationStrings = {
'costs.receipts': 'Hóa đơn',
'costs.receiptsTitle': 'Hóa đơn & Biên lai',
'costs.attachReceipt': 'Đính kèm hóa đơn / biên lai',
+ 'costs.scan.button': 'Quét hóa đơn',
+ 'costs.scan.reading': 'Đang đọc hóa đơn…',
+ 'costs.scan.review': 'Xem lại khoản chi',
+ 'costs.scan.nothingRead': 'Không đọc được hóa đơn nào từ ảnh này.',
+ 'costs.scan.failed': 'Không thể bắt đầu đọc hóa đơn.',
+ 'costs.scan.title': 'Quét hóa đơn',
+ 'costs.scan.accepted': 'Ảnh của một hóa đơn (JPG, PNG hoặc WEBP, tối đa 10 MB). Mô hình AI đọc ảnh ở chế độ nền, sau đó bạn kiểm tra khoản chi trước khi lưu.',
+ 'costs.scan.dropHere': 'Thả ảnh hóa đơn vào đây, hoặc bấm để chọn hay chụp ảnh',
+ 'costs.scan.dropActive': 'Thả ảnh để quét',
+ 'costs.scan.start': 'Quét',
+ 'costs.scan.unsupported': 'Chỉ có thể quét ảnh: JPG, PNG hoặc WEBP.',
'costs.noReceipts': 'Chưa đính kèm hóa đơn nào',
'costs.deleteReceipt': 'Xóa hóa đơn',
'costs.viewReceipt': 'Xem hóa đơn',
diff --git a/shared/src/i18n/vi/reservations.ts b/shared/src/i18n/vi/reservations.ts
index 631369a7c..04a2dcb80 100644
--- a/shared/src/i18n/vi/reservations.ts
+++ b/shared/src/i18n/vi/reservations.ts
@@ -130,6 +130,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropActive': 'Thả tập tin để nhập',
'reservations.import.acceptedFormats':
'Đã chấp nhận: EML, PDF, PKPass, HTML, TXT (tối đa 10 MB mỗi cái, tối đa 5 tệp)',
+ 'reservations.import.acceptedPhotos': 'Ảnh (JPG, PNG, WEBP) được mô hình AI đọc.',
'reservations.import.parsing': 'Đang phân tích tệp…',
'reservations.import.previewHeading': '{count} đã tìm thấy (các) đặt chỗ',
'reservations.import.previewEmpty': 'Không thể trích xuất đặt chỗ nào từ các tập tin đã tải lên.',
diff --git a/shared/src/i18n/vi/settings.ts b/shared/src/i18n/vi/settings.ts
index 9b00d7c48..f40a8a91d 100644
--- a/shared/src/i18n/vi/settings.ts
+++ b/shared/src/i18n/vi/settings.ts
@@ -399,9 +399,9 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': 'Nơi mô hình chạy — máy chủ Ollama cục bộ hoặc endpoint tương thích với OpenAI.',
'settings.aiParsing.apiKey': 'Khóa API',
'settings.aiParsing.apiKeyHint': 'Được lưu trữ mã hóa. Để trống để giữ khóa hiện tại.',
- 'settings.aiParsing.multimodal': 'Gửi tài liệu dưới dạng hình ảnh',
+ 'settings.aiParsing.multimodal': 'Mô hình đọc được hình ảnh',
'settings.aiParsing.multimodalHint':
- 'Dành cho các mô hình hỗ trợ thị giác — gửi PDF gốc thay vì văn bản đã trích xuất.',
+ 'Bật cho mô hình đọc được hình ảnh để có thể nhập hoặc quét ảnh.',
'settings.aiParsing.toast.saved': 'Đã lưu cài đặt AI',
'settings.aiParsing.toast.saveError': 'Không thể lưu cài đặt AI',
'settings.tabs.appearance': 'Giao diện',
diff --git a/shared/src/i18n/zh-TW/admin.ts b/shared/src/i18n/zh-TW/admin.ts
index 9e1172adc..e48e2ed36 100644
--- a/shared/src/i18n/zh-TW/admin.ts
+++ b/shared/src/i18n/zh-TW/admin.ts
@@ -620,6 +620,11 @@ const admin: TranslationStrings = {
'從各位使用者自行連接的 Dawarich 執行個體讀取造訪紀錄與記錄的路線',
'admin.addons.catalog.llm_parsing.name': 'AI 解析',
'admin.addons.catalog.llm_parsing.description': '用你選擇的 AI 模型讀取內建解析器無法識別的訂位',
+ 'admin.addons.llm.vision.auto': '自動',
+ 'admin.addons.llm.vision.on': '是',
+ 'admin.addons.llm.vision.off': '否',
+ 'admin.addons.llm.vision.hintLocal': '「自動」會詢問 Ollama 伺服器此模型能否讀取圖片。',
+ 'admin.addons.llm.vision.hintCloud': '對雲端模型而言，「自動」即為「否」。若此模型能讀取圖片，請選擇「是」。',
'admin.addons.enabled': '已啟用',
'admin.addons.disabled': '已停用',
'admin.addons.type.trip': '旅行',
diff --git a/shared/src/i18n/zh-TW/budget.ts b/shared/src/i18n/zh-TW/budget.ts
index fcaf6ec55..366aab5fc 100644
--- a/shared/src/i18n/zh-TW/budget.ts
+++ b/shared/src/i18n/zh-TW/budget.ts
@@ -153,6 +153,17 @@ const budget: TranslationStrings = {
'costs.receipts': '收據',
'costs.receiptsTitle': '收據與發票',
'costs.attachReceipt': '附加收據/發票',
+ 'costs.scan.button': '掃描收據',
+ 'costs.scan.reading': '正在讀取收據…',
+ 'costs.scan.review': '檢查支出',
+ 'costs.scan.nothingRead': '無法從這張照片讀取收據。',
+ 'costs.scan.failed': '無法開始讀取收據。',
+ 'costs.scan.title': '掃描收據',
+ 'costs.scan.accepted': '一張收據的照片（JPG、PNG 或 WEBP，最大 10 MB）。AI 模型會在背景讀取，儲存前你可以檢查這筆支出。',
+ 'costs.scan.dropHere': '將收據照片拖到這裡，或點擊選擇或拍攝照片',
+ 'costs.scan.dropActive': '放開以掃描照片',
+ 'costs.scan.start': '掃描',
+ 'costs.scan.unsupported': '只能掃描照片：JPG、PNG 或 WEBP。',
'costs.noReceipts': '未附加收據',
'costs.deleteReceipt': '刪除收據',
'costs.viewReceipt': '查看收據',
diff --git a/shared/src/i18n/zh-TW/reservations.ts b/shared/src/i18n/zh-TW/reservations.ts
index 4e1b2d0b5..59a9d58de 100644
--- a/shared/src/i18n/zh-TW/reservations.ts
+++ b/shared/src/i18n/zh-TW/reservations.ts
@@ -128,6 +128,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': '將訂位確認檔案拖放到此處，或點擊選擇',
'reservations.import.dropActive': '放開檔案以匯入',
'reservations.import.acceptedFormats': '支援格式：EML、PDF、PKPass、HTML、TXT（每個最大 10 MB，最多 5 個檔案）',
+ 'reservations.import.acceptedPhotos': '照片（JPG、PNG、WEBP）由 AI 模型讀取。',
'reservations.import.parsing': '正在解析檔案…',
'reservations.import.previewHeading': '找到 {count} 筆預訂',
'reservations.import.previewEmpty': '無法從上傳的檔案中提取任何預訂資訊。',
diff --git a/shared/src/i18n/zh-TW/settings.ts b/shared/src/i18n/zh-TW/settings.ts
index 509f2b0b5..a9a21bbc1 100644
--- a/shared/src/i18n/zh-TW/settings.ts
+++ b/shared/src/i18n/zh-TW/settings.ts
@@ -372,8 +372,8 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': '模型執行的位置 — 本機 Ollama 伺服器，或相容於 OpenAI 的端點。',
'settings.aiParsing.apiKey': 'API 金鑰',
'settings.aiParsing.apiKeyHint': '以加密方式儲存。留空即可保留目前的金鑰。',
- 'settings.aiParsing.multimodal': '以圖片形式傳送文件',
- 'settings.aiParsing.multimodalHint': '適用於具備視覺能力的模型 — 傳送原始 PDF，而非擷取出的文字。',
+ 'settings.aiParsing.multimodal': '模型可讀取圖片',
+ 'settings.aiParsing.multimodalHint': '適用於能讀取圖片的模型，開啟後即可匯入或掃描照片。',
'settings.aiParsing.toast.saved': 'AI 設定已儲存',
'settings.aiParsing.toast.saveError': '無法儲存 AI 設定',
'settings.tabs.appearance': 'Appearance',
diff --git a/shared/src/i18n/zh/admin.ts b/shared/src/i18n/zh/admin.ts
index 2873b3ea4..acd759103 100644
--- a/shared/src/i18n/zh/admin.ts
+++ b/shared/src/i18n/zh/admin.ts
@@ -577,6 +577,11 @@ const admin: TranslationStrings = {
'从每位使用者自行连接的 Dawarich 实例读取到访记录和记录的路线',
'admin.addons.catalog.llm_parsing.name': 'AI 解析',
'admin.addons.catalog.llm_parsing.description': '用你选择的 AI 模型读取内置解析器无法识别的预订',
+ 'admin.addons.llm.vision.auto': '自动',
+ 'admin.addons.llm.vision.on': '是',
+ 'admin.addons.llm.vision.off': '否',
+ 'admin.addons.llm.vision.hintLocal': '“自动”会询问 Ollama 服务器该模型能否读取图像。',
+ 'admin.addons.llm.vision.hintCloud': '对云端模型，“自动”即为“否”。如果该模型能读取图像，请选择“是”。',
'admin.addons.enabled': '已启用',
'admin.addons.disabled': '已禁用',
'admin.addons.type.trip': '旅行',
diff --git a/shared/src/i18n/zh/budget.ts b/shared/src/i18n/zh/budget.ts
index aea929eef..56ac04360 100644
--- a/shared/src/i18n/zh/budget.ts
+++ b/shared/src/i18n/zh/budget.ts
@@ -153,6 +153,17 @@ const budget: TranslationStrings = {
'costs.receipts': '收据',
'costs.receiptsTitle': '收据与发票',
'costs.attachReceipt': '附加收据/发票',
+ 'costs.scan.button': '扫描收据',
+ 'costs.scan.reading': '正在读取收据…',
+ 'costs.scan.review': '检查支出',
+ 'costs.scan.nothingRead': '无法从这张照片中读取收据。',
+ 'costs.scan.failed': '无法开始读取收据。',
+ 'costs.scan.title': '扫描收据',
+ 'costs.scan.accepted': '一张收据的照片（JPG、PNG 或 WEBP，最大 10 MB）。AI 模型在后台读取，保存前你可以检查这笔支出。',
+ 'costs.scan.dropHere': '将收据照片拖到这里，或点击选择或拍摄照片',
+ 'costs.scan.dropActive': '松开以扫描照片',
+ 'costs.scan.start': '扫描',
+ 'costs.scan.unsupported': '只能扫描照片：JPG、PNG 或 WEBP。',
'costs.noReceipts': '未附加收据',
'costs.deleteReceipt': '删除收据',
'costs.viewReceipt': '查看收据',
diff --git a/shared/src/i18n/zh/reservations.ts b/shared/src/i18n/zh/reservations.ts
index 67c0199e0..283a94642 100644
--- a/shared/src/i18n/zh/reservations.ts
+++ b/shared/src/i18n/zh/reservations.ts
@@ -128,6 +128,7 @@ const reservations: TranslationStrings = {
'reservations.import.dropHere': '将预订确认文件拖放到此处，或点击选择',
'reservations.import.dropActive': '松开文件以导入',
'reservations.import.acceptedFormats': '支持格式：EML、PDF、PKPass、HTML、TXT（每个最大 10 MB，最多 5 个文件）',
+ 'reservations.import.acceptedPhotos': '照片（JPG、PNG、WEBP）由 AI 模型读取。',
'reservations.import.parsing': '正在解析文件…',
'reservations.import.previewHeading': '找到 {count} 个预订',
'reservations.import.previewEmpty': '无法从上传的文件中提取任何预订信息。',
diff --git a/shared/src/i18n/zh/settings.ts b/shared/src/i18n/zh/settings.ts
index 06ddbfd51..5c63c44a2 100644
--- a/shared/src/i18n/zh/settings.ts
+++ b/shared/src/i18n/zh/settings.ts
@@ -369,8 +369,8 @@ const settings: TranslationStrings = {
'settings.aiParsing.baseUrlHint': '模型运行的位置——本地 Ollama 服务器或兼容 OpenAI 的接口。',
'settings.aiParsing.apiKey': 'API 密钥',
'settings.aiParsing.apiKeyHint': '加密存储。留空则保留当前密钥。',
- 'settings.aiParsing.multimodal': '以图像形式发送文档',
- 'settings.aiParsing.multimodalHint': '适用于支持视觉的模型——发送原始 PDF 而非提取的文本。',
+ 'settings.aiParsing.multimodal': '模型可读取图像',
+ 'settings.aiParsing.multimodalHint': '适用于能读取图像的模型，开启后可导入或扫描照片。',
'settings.aiParsing.toast.saved': 'AI 设置已保存',
'settings.aiParsing.toast.saveError': '无法保存 AI 设置',
'settings.tabs.appearance': '外观',
diff --git a/shared/src/index.ts b/shared/src/index.ts
index e7d2aa19c..b25a319da 100644
--- a/shared/src/index.ts
+++ b/shared/src/index.ts
@@ -29,6 +29,7 @@ export * from './vacay/vacay.schema';
export * from './packing/packing.schema';
export * from './todo/todo.schema';
export * from './budget/budget.schema';
+export * from './budget/receipt-scan.schema';
export * from './reservation/reservation.schema';
export * from './reservation/ki-reservation.schema';
export * from './datetime/datetime-normalize';
diff --git a/wiki/AI-Booking-Import.md b/wiki/AI-Booking-Import.md
index b1be546a2..9f5e8f859 100644
--- a/wiki/AI-Booking-Import.md
+++ b/wiki/AI-Booking-Import.md
@@ -26,7 +26,9 @@ The addon supports three providers:
| **OpenAI** | OpenAI's API, or any **OpenAI-compatible** endpoint via a custom base URL | Needs an API key. |
| **Anthropic** | Anthropic's API | Needs an API key. **Reads PDFs — including scans — natively.** |
-> **Scanned PDFs:** Local and OpenAI-compatible models receive the document's *extracted text*. A scanned or image-only PDF has no text layer, so those providers return nothing for it. Only **Anthropic** ingests the raw PDF and can read scans.
+> **Scanned PDFs:** Local and OpenAI-compatible models receive the document's *extracted text*. A scanned or image-only PDF has no text layer: when the model reads images, its first two pages are drawn as images and sent like a photo; otherwise those providers return nothing for it. **Anthropic** ingests the raw PDF and reads scans either way.
+>
+> **Photos:** a photo (JPG, PNG, WEBP) is sent to the model as an image, with every provider, when the model reads images. See [Photos](#photos).
## Admin: instance-wide configuration
@@ -38,6 +40,7 @@ When you enable the addon, a configuration panel appears directly under it in [A
- **Base URL** — shown for every provider except Anthropic. Defaults to `http://localhost:11434/v1` for a local Ollama server, or `https://api.openai.com/v1` for OpenAI. Point it at any OpenAI-compatible endpoint here.
- **API key** — optional for a local server (`(often not required)`), required for the cloud providers. Stored **encrypted**; it is shown masked (`••••••••`) once saved, and leaving it unchanged keeps the stored key.
- **Model** — the model id (e.g. `qwen3.5:4b`, `gpt-4o`, `claude-opus-4-8`).
+- **Model reads images** — **Automatic**, **Yes** or **No**. With the Local provider, **Automatic** asks the Ollama server what the model can do (`/api/show`) and follows its answer; with OpenAI or Anthropic, **Automatic** means no. **Yes** and **No** are applied as set, whatever the server or the model id says. Photos are only offered when this resolves to yes (see [Photos](#photos)).
If you set a provider and model here, it applies to **all users** and overrides their personal settings. Leave the panel blank to let each user bring their own model (see below).
@@ -46,7 +49,7 @@ If you set a provider and model here, it applies to **all users** and overrides
With the **Local** provider selected, the panel manages your Ollama server directly:
- **Installed on the server** lists the models Ollama already has, with a **Refresh** button. Click a model to select it.
-- **Pull a recommended model** downloads a model with a live progress bar. The one recommended model is **Qwen3.5 — 4B** (`qwen3.5:4b`, 3.4 GB, 256K context): small and quick on a CPU-only host, "thinking" is disabled automatically, Apache-2.0. Like every local model it receives the document's extracted text, so scanned PDFs still need Anthropic (see below). Once the pull finishes it is selected automatically.
+- **Pull a recommended model** downloads a model with a live progress bar. The one recommended model is **Qwen3.5 — 4B** (`qwen3.5:4b`, 3.4 GB, 256K context): small and quick on a CPU-only host, "thinking" is disabled automatically, Apache-2.0. It reads images, so with **Model reads images** on Automatic it takes photos. A PDF with a text layer reaches it as extracted text; a scanned one as its first two pages drawn as images (see above). Once the pull finishes it is selected automatically.
You can also select any other model already installed on the server, or type a model id by hand.
@@ -58,7 +61,7 @@ If an admin leaves the instance config blank, each user can configure their own
The fields are a **Provider** (only **OpenAI** or **Anthropic** here), a **Model** id, and an **API key** that is *stored encrypted* (leave blank to keep the current key). There is no personal Base URL: the address this server calls is instance configuration, so a local (Ollama) model can only be set up by an admin on the addon, and the server answers 403 to anyone, admins included, who tries to store a personal base URL or a personal `local` provider.
-There is also a **Send documents as images** toggle. It is stored per user, but extraction currently ignores it: only Anthropic is sent the raw PDF, every other provider always gets the extracted text.
+There is also a **Model reads images** toggle. On, photos are offered and sent to the personal model as images; off, they are not offered. PDFs are unaffected: only Anthropic is sent the raw PDF, every other provider gets the extracted text.
> **Precedence:** an admin instance model always wins. Personal settings only take effect when no instance-wide model is configured.
@@ -67,7 +70,7 @@ There is also a **Send documents as images** toggle. It is stored per user, but
The upload flow is the normal booking import — the AI simply runs behind it:
1. In the trip planner, open the **Bookings** tab and click **Import from file**.
-2. Drop your files (EML, PDF, PKPass, HTML, TXT — up to 5 files, 10 MB each) onto the upload area.
+2. Drop your files (EML, PDF, PKPass, HTML, TXT — up to 5 files, 10 MB each) onto the upload area. When the model reads images, photos (JPG, PNG, WEBP) are accepted too.
3. The upload dialog closes right away and a **background widget** (bottom-right) shows *Parsing files…* with a running count. You can keep navigating TREK while it works; the widget survives a page reload and even follows you to other pages.
4. When parsing finishes, click the widget's **Import** button to start the review.
5. Each parsed booking opens **pre-filled in the normal reservation (or transport) editor**, one at a time. Nothing is saved until you confirm each one.
@@ -88,12 +91,24 @@ The model is asked to capture the full booking — including **every leg of a mu
- **Linked cost** — if the [Costs/Budget addon](Budget-Tracking) is enabled and the booking has a price, a linked expense is created. Without that addon, the price stays on the reservation only.
- **Source document** — the uploaded file is attached to the reservation's files.
+## Photos
+
+When the configured model reads images (see **Model reads images** above), the import accepts photos as well: a phone picture of a paper train ticket, a boarding pass or a hotel confirmation. The upload dialog then lists *Photos (JPG, PNG, WEBP) are read by the AI model.*, and a photo goes through the same background job, review and linked cost as any other file.
+
+- A photo always goes to the model: KDE Itinerary is not asked, since it reads documents, not pictures.
+- A scanned PDF (no text layer) is read the same way: its first two pages are drawn as images and sent to the model.
+- It is shrunk to 1600 px on its longest side before it is sent. A model pays for every pixel and does not read a till roll better at twelve megapixels; on a local `qwen3.5:4b` running on CPU the same ticket took several times longer at full size.
+- HEIC is not accepted. A phone's photo picker hands over a JPEG.
+- When the model does not read images, the dialog does not offer photos, and the server refuses one sent anyway with *The configured AI model does not read photos*.
+
+Photographing a receipt for the Costs tab is covered in [Budget-Tracking](Budget-Tracking#scanning-a-receipt).
+
## Good to know
- **No manual migration**, and the addon is configured in the UI. The one environment variable it reads is `LLM_TIMEOUT_MS` (see [Environment-Variables](Environment-Variables)).
- **Local inference can be slow.** On a CPU-only host a single booking can take tens of seconds to a couple of minutes; TREK allows a model 15 minutes per document by default, which `LLM_TIMEOUT_MS` raises or lowers. Uploads are parsed **one at a time** per user, so several files queue rather than run in parallel.
- **Parse jobs are kept for about 10 minutes** after they finish. Start the review within that window.
-- **Privacy** — with the Local provider nothing leaves your network. With OpenAI or Anthropic, the document's text (or, for Anthropic, the PDF itself) is sent to that provider for extraction.
+- **Privacy** — with the Local provider nothing leaves your network. With OpenAI or Anthropic, the document's text (or, for Anthropic, the PDF itself, and for either, a photo) is sent to that provider for extraction.
- **API keys are never returned in plaintext** — they are encrypted at rest and only ever shown masked.
## Related pages
diff --git a/wiki/Budget-Tracking.md b/wiki/Budget-Tracking.md
index 030b3eb2b..29c397369 100644
--- a/wiki/Budget-Tracking.md
+++ b/wiki/Budget-Tracking.md
@@ -78,6 +78,17 @@ A row with receipts shows a **Receipts** chip beside the name, with the count wh
**Remove receipt** in the editor only unlinks the file from the expense. The file itself stays on the trip, because editing an expense (`budget_edit`) does not carry the file permission: to get rid of the file, delete it in the Files tab, which needs `file_delete`. Uploading a receipt goes through the trip's file upload, so it needs `file_upload` on top of `budget_edit`. A file that is also linked to a place or a booking keeps those links. See [Documents-and-Files](Documents-and-Files).
+### Scanning a receipt
+
+With the [AI Parsing](AI-Booking-Import) addon on and a model that reads images (see *Model reads images* there), a **Scan receipt** button sits beside **Add expense**, and on a phone the Costs header carries a scan icon. It is offered to whoever may add expenses.
+
+1. Click **Scan receipt**. A dialog like the booking import's takes one photo of the receipt (JPG, PNG or WEBP, up to 10 MB): drop it on the box, or click the box to pick one or, on a phone, take one. Then click **Scan**.
+2. The dialog closes and the model reads the photo in the background: the **background tasks** widget shows *Reading the receipt…*, and you can keep using TREK meanwhile. On a local model running on CPU this takes from a few seconds to a couple of minutes.
+3. When it is done, **Review expense** opens the expense editor pre-filled with what was read: the merchant as the name, the total, the currency and the day. The split stays **Equally**; switch it to **Ticket** and the receipt's lines are already listed, each shared by everyone. The photo is already in **Receipts & Invoices**, uploaded when you save like any receipt you attach yourself, which needs `file_upload` too.
+4. Check everything, pick the category, who paid and the split, and save. Nothing is stored before that.
+
+The scan reads one photo per expense and does not guess the category. A photo nothing could be read from ends on *No receipt could be read from this photo.*, with the reason under it. Scan jobs are kept for about 10 minutes after they finish, like [booking imports](AI-Booking-Import#good-to-know).
+
## Who paid
**Who paid?** in the expense editor records who actually put the money down. It is the other half of the settlement maths — the split says who owes for the expense, this says who is out of pocket for it: