- Block the list-cover upload in demo mode (mirror the trips cover endpoint).
- Restrict list/place links to http(s) (schema) and normalise scheme-less URLs
to https:// on save, so a bare "booking.com" no longer resolves as a relative
SPA route (and javascript:/data: hrefs are rejected).
- Place detail: surface save errors with a toast instead of silently swallowing
a 400 and leaving the sheet stuck in edit mode.
- List editor: don't create a duplicate list when a retry follows a cover-upload
failure (reuse the created id); revoke the cover preview object URL.
- Map controls: one top bar (left toggle/select, right add/search) so they can't
overlap on a narrow split map — the search shrinks instead.
- Dashboard list badge: full-opacity colour wash so the name stays legible over
bright covers.
Dashboard widget (B): the collections tool now shows the user's LISTS as
compact colour-washed badges (cover image tinted with the list colour, or a
gradient) that jump to the list — one list() call, no N+1.
List details (C): lists gain a description, a custom cover image (uploaded to
/uploads/covers, tinted with the list colour in the hero) and links. A shared
ListEditorModal handles both create and edit; the hero shows the description +
link chips. New `links` JSON column on collections + collection_places
(migration 151) with parse/serialize in the service; a POST :id/cover upload
endpoint mirroring trips; cover-file cleanup path-confined locally.
Place detail (D): clicking a place opens a bottom sheet (no backdrop, so the
map stays visible) — status cycle, copy-to-trip, remove, and an edit mode with
a markdown description + links editor (collectionsApi.updatePlace, now wired
via a store action). A "+" next to the search adds a place to the list via the
maps search.
Fusion + fixes (E): the owner can now remove an accepted member (kick) — new
removeMember service/route/store + a button in ShareCollectionModal, with a
collections:removed WS bounce. findMembership no longer matches on name alone
(coordinate proximity required, killing "Starbucks everywhere" false positives).
loadCollection swallows a 403/404 after a leave/remove so the URL sync can't
throw uncaught. Grid remnants gone; the map select toggle moved onto the map.
New strings across all 22 locales; i18n parity strict passes.
Adds the Collections addon backend: a server-wide-per-user library of saved
places, independent of any trip, with multiple named lists, an idea/want/visited
status, and Vacay-style fusion invitations to share a list with other users.
- Data: collection / collection_members / collection_places / collection_place_tags
tables (+ migration and baseline schema). Saved places carry the owner plus a
nullable saved_by so a member deleting their account can't drop shared content.
- Service: list + place CRUD with owner-or-accepted-member visibility, dedup,
status, save-from-trip and copy-to-trip (reusing the trip copy column list),
and the full fusion-invitation state machine mirrored from vacay (send / accept
/ decline / cancel / leave) with a websocket broadcast and an invite
notification. Deleting a list snapshots its members and notifies them.
- NestJS module + addon guard (404 before auth), registered in the app module.
- Widens the place photo cache reference check to count collection places so the
nightly sweep no longer evicts photos a saved place still uses.
- collection_invite notification wired across all 22 locales.
Rework the private-packing flag into a full sharing model. Every item is now
Common (the group pool — where all existing items live, so nothing breaks),
Personal (private to its owner) or Shared with specific people (it shows up on
those travelers' own lists, marked "by <bringer>"). is_private discriminates
restricted from common; a new packing_item_recipients table holds who a shared
item covers, and packing_item_contributors records "I can bring that too"
pledges on Common items.
The panel gains a Gemeinsam / Meine Liste view switch, each item a sharing
control (owner sets the tier + the people it covers), and Common items can be
co-brought or cloned onto your personal list. Visibility is enforced server-side
in listItems and the WebSocket broadcasts are scoped to exactly who can see an
item across every tier transition. All 22 locales stay in parity.
Add "guest" trip participants — people without a Trek account who can still be
assigned to costs, packing, to-dos and day-plan activities. A guest is a
credential-less users row (is_guest=1) joined into trip_members, so it is
assignable everywhere a real member is, with the cost-splitting, settlement,
packing and assignment paths working unchanged.
Guests are firewalled from everything account-related: they can never sign in
(password, OIDC and reset lookups skip them), never appear in the global user
directory, the member-add picker or admin user management, are never resolved as
notification recipients, can't be invited to another trip, and can't be made
owner. The trip owner manages guests from the share dialog in a dedicated,
clearly-labelled section (add / rename / remove), and guests carry a "Guest"
badge wherever members are picked. All 22 locales stay in parity.
Add POST /api/trips/:id/transfer so the owner can hand a trip to one of its
existing members. The swap runs in a transaction: the new owner takes
trips.user_id and the former owner is kept on as a regular member, so nobody
loses access. The endpoint is owner-only, writes a trip.transfer_ownership
audit entry and broadcasts the refreshed trip. The members modal gains a
"Make owner" action, shown only to the current owner.
Add drag-to-reorder to the packing and to-do lists, mirroring the budget
panel's native HTML5 drag pattern. A drag within a filtered/grouped view is
mapped back onto the global order so untouched items keep their place, and the
order persists optimistically via the existing reorder endpoints.
Packing items can now be marked private (#858): a private item is visible only
to its owner. createItem/bulkImport stamp the owner, listItems filters by the
viewer, and the WebSocket broadcasts are scoped to the owner so a private item
never reaches another member's screen — including the public/private toggle
transitions. Owners get a lock toggle and a private indicator on their items.
Markdown (.md/.markdown) is now an allowed upload type and opens in a rendered preview in the file manager instead of just downloading. Reuses the existing react-markdown stack with rehype-sanitize (these are untrusted uploads, so output is sanitized) and detects markdown by extension first since browsers send unreliable MIME for .md.
Bookings get a first-class url column (migration) instead of users pasting links into notes. It's editable in the booking modal and rendered as a clickable link on the reservation card. The reservation request schemas are open passthroughs, so only the entity schema + service SQL enumerate it.
The gallery upload input now accepts image/*,video/* — update the two JourneyDetailPage selectors that matched the old value. The files/journey e2e suites mock fileService and were missing the new MAX_VIDEO_SIZE / isVideoExtension / isVideoMime exports, which broke module load.
Security: the gallery-video poster is now always stored as .jpg instead of the client-supplied extension, so a poster declared image/* but named x.html / x.js can't be written with that extension and served inline same-origin; local gallery files are also served with X-Content-Type-Options: nosniff.
Robustness: rejected/unauthorised uploads no longer orphan their bytes on disk (the gallery-video and file-manager handlers unlink before throwing); the file-manager per-type size cap is keyed on the extension like the filter, so a real video labelled application/octet-stream isn't wrongly rejected. UX: the file-manager thumbnail strip shows a play placeholder for video instead of a broken image; shared (public) journeys now return media_type and play videos with a play badge; and a poster-less video shows a neutral tile instead of a broken thumbnail.
The file manager (which already attaches files to a place/activity) now accepts video uploads up to the larger video cap — other types stay at the document limit — and the lightbox plays them with the Plyr player over the plain same-origin download URL, so cookie auth and HTTP Range both work. Videos are excluded from the offline blob prefetch so one clip can't evict a trip's documents.
Immich timeline and album listings no longer filter out videos; each asset now carries its media type, which the provider picker forwards when linking. A linked video streams through Immich's transcoded /video/playback endpoint, and the asset proxy forwards the viewer's Range header (and passes 206/Content-Range back) so the player can seek. Synology video stays excluded until its stream API is verified.
Adds media_type/media_types to the provider-photos request contract.
Picking a video in the journey gallery now captures a poster frame + duration in the browser and uploads the raw clip; the grid shows the poster with a play badge and the lightbox plays it with a native video player (HTTP Range seeking). Images keep their existing HEIC-normalised path. No server-side transcoding.
Server media_type work was committed separately.
trek_photos gains a media_type column (migration) so the registry can hold video as well as images. A new POST :id/gallery/video endpoint accepts a video plus a client-captured poster (500 MB cap, video MIME/extension allowlist), stores the poster as the thumbnail, and the photo stream serves the poster for the thumbnail kind and the raw file (HTTP Range) for the original — without running the image thumbnailer on video bytes.
Update handlers accept an optional X-Base-Updated-At token and reject a stale overwrite with 409, returning the current server row. An absent token keeps the existing last-write-wins behaviour, so older clients are unaffected. packing_items gains an updated_at column (migration + stamped on every insert) so it can take part in conflict detection too.
Closes the UI half of #1168: in the Places selection mode, a new tag button
before delete opens a category picker that applies one category (or "No
category") to every selected place in a single request. Adds a REST
/places/bulk-update endpoint reusing updatePlacesMany, an offline-aware repo +
store action that patches both the place pool and the day-assignment
projections, undo grouped by each place's prior category, and the i18n keys
across all locales.
Apply the same field values to many places in one call instead of one
update_place per place — e.g. re-categorising 80 POIs at once. Adds the
updatePlacesMany service (one transaction, trip-scoped, partial patch
built on updatePlace) and the bulk_update_places MCP tool with the usual
demo/access/place_edit guards and a place:updated broadcast per place.
- Resolve feed URLs against the request host when APP_URL is unset, so the
webcal:// / Add-to-Google links work on a default install (not just behind a
configured reverse proxy).
- Give the public link a real off switch: POST enables, PUT rotates, DELETE
clears the token (feed_token = NULL). The subscribe dialog no longer mints a
token just from being opened — the user opts in explicitly.
- Fold ICS content lines at 75 octets (UTF-8 safe) in exportICS, so download
and feed both stay RFC 5545-compliant for long/non-ASCII summaries.
- Extract VEVENTs by structural line scan instead of a lazy END:VEVENT regex
that user text could truncate.
- URL-encode the Google Calendar cid; mirror feed_token into schema.ts.
- Collapse the duplicated all-trips modal into the shared IcsSubscribeModal.
Adds TripIt-style live calendar subscriptions alongside the existing one-time
.ics download. A trip (or all of a user's trips) exposes a secret, revocable
feed URL that Google/Apple/Outlook poll to stay in sync.
- Public read endpoints GET /api/feed/trip/:token.ics and /api/feed/user/:token.ics
(no auth — the secret token is the credential), reusing the existing exportICS()
generator and adding REFRESH-INTERVAL / X-PUBLISHED-TTL hints.
- JWT-guarded token endpoints to generate (lazy, idempotent) and regenerate/revoke
per-trip and per-user feed tokens; tokens stored in nullable feed_token columns.
- All-trips feed excludes archived trips and trips ended >90 days ago.
- UI: ICS toolbar button becomes a Download/Subscribe menu; modal offers one-click
"Add to Google Calendar" (render?cid=webcal://) and a webcal:// link for
Apple/Outlook, plus copy-link fallbacks. All-trips feed reachable from dashboard.
- Feed base URL read from the existing APP_URL env var.
Purely additive: new endpoints + two nullable columns, no breaking changes.
Tests: server/tests/e2e/feeds.e2e.test.ts covers lazy token generate + idempotency,
regenerate-invalidates-old, 401/404 auth+access, public feed content-type + hint
injection, unknown-token 404, and the archived/>90-day all-trips exclusion.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add per-payer and per-member custom split amounts with Equally, Custom and
Ticket split modes on top of the existing equal split, keep legacy "paid by"
expenses working, and document the modes in the Budget Tracking wiki page.
When the server issues a Secure session cookie but the request arrived over
plain HTTP (the common LAN install over http://ip:3000), the browser drops
the cookie and the next request dead-ends on a bare "Access token required" —
the top source of avoidable install issues. The login response now flags this
exact case and the login page shows a localized box explaining the fix (use
HTTPS, or set COOKIE_SECURE=false) with a link to the Troubleshooting guide.
It only triggers in the real failure case, never for correct HTTPS setups.
Add a Help section (profile menu, /help) that renders the GitHub wiki inside
TREK. /api/help fetches the wiki markdown — the nav from _Sidebar.md, pages,
and proxied images — from GitHub and caches it (1h TTL, serves stale on
outage), so it auto-syncs on wiki edits with no redeploy and the client never
calls GitHub directly. The page is styled to match TREK with a section
sidebar, search and react-markdown; wiki [[links]] are rewritten to in-app
routes and HTML-comment placeholders are stripped. Page state lives in a
useHelp() hook per the page pattern. Adds nav.help and a help namespace
across all locales.
Shared AppearanceConfig (color scheme, accent, transparency, per-tier type scale, density, reduce-motion and per-device dashboard widgets) stored as one JSON blob under the existing settings key. normalizeAppearance never throws, so a malformed/partial/future blob degrades to the neutral default and can never reach the DOM. No DB migration; the default reproduces today's look exactly.
Adds unit coverage for saveUnsplashCover (host check, content-type
and size limits, download failure), the searchUnsplashPhotos error
and success paths, and the PUT handler internalising a hot-link.
Updates the existing PUT tests for the now-async handler.
Previously a selected Unsplash photo was stored as a remote
images.unsplash.com hot-link, so covers broke offline and on link
rot. The trip PUT handler now fetches the picked image through the
SSRF guard and saves it under uploads/covers, rewriting cover_image
to the local path (502 if the download fails). Also debounces the
cover search so a slow earlier request can no longer overwrite newer
results, drops a dead userId parameter, and reverts an unrelated
vite proxy change.
The mapper read only `departureScheduled`/`arrivalScheduled`, but those columns
are optional in AirTrail and stay null for manually-entered flights — where
`departure`/`arrival` are the only times set. So the import dropped the departure
clock (date-only) and the whole arrival (no date, no time), exactly as reported.
AirTrail's own rule is "use departure if available, otherwise fall back to
departureScheduled". Mirror that: prefer the scheduled instant, fall back to the
primary departure/arrival, in mapFlightToReservation, normalizeFlight, and the
sync hash. Hashing the resolved instant means flights already imported without a
scheduled time re-sync once and pick up their clock automatically; flights that
do have scheduled times are unaffected (no spurious re-sync).
Tests: 3 new mapper cases (fallback mapping, picker preview, hash tracking);
two existing cases that asserted the scheduled-only behaviour updated to the
"neither time set" case. Full server suite green (4085).
This branch added a clamp-to-nearest-day fallback to resolveDayIdFromTime so an
imported booking whose exact date has no day row still lands on a day. After
rebasing onto dev, that collided with #1288's resyncReservationDays, which
relies on the original "null when no exact day" semantics to leave a booking
whose date now falls outside the range untouched — instead it snapped to an edge
day (TRIP-SVC-019 failed: expected day_id kept, got the clamped one).
Make clampToNearest an opt-in parameter (default true, preserving the import
behaviour for create/update) and have resyncReservationDays pass false, so
out-of-range bookings keep their day_id. Full server suite green (4082).
The new LLM extraction router shipped with little branch coverage, dropping src/nest below the 80% gate. Add unit tests for routeExtraction (flights/single/union/error paths, deterministic booking-wide fill), the native Ollama format client, the provider factory, the local-router service path with its type-aware text cap, the flat->schema.org mapper's remaining reservation types, and the background import-jobs runner. Also remove the now-unused validate.ts (only its FlatLike type was still referenced; moved to flat-schemas).
A long single-booking PDF (e.g. an 11-page rental voucher) spent ~200s on CPU prompt-eval at the 16k cap, though its data sits in the first ~2k. Cap non-flight docs at 6k (flights keep 16k for all legs). Also make the rental operator a required field so the car gets a real title.
After dropping the vendor templates, the model skipped the (often unlabeled) Expedia-style hotel address — making address a required schema field forces it to emit the street-address line, restoring the booking's location/place. Also hint the rental company so a car booking gets a real title instead of the generic fallback.
Now that a capable instruct model (Qwen3-8B, thinking off) reads name/address/dates/legs reliably across formats, the per-vendor template short-circuit distorted more than it fixed: brittle on layout variations and overriding the better model output. Remove the template layer; the model extracts the structure and Schicht 2 backfills the confirmation/total and takes the currency from the document's own symbol (correcting model misreads like ¥→$). Per-type prompts now also ask for address and price/currency.
Hybrid/reasoning models (Qwen3 and similar) default to emitting reasoning tokens, which collide with Ollama's format-grammar constraint — on CPU this produced null/unparseable output and blew the latency budget (qwen3:8b: null or 300s timeouts vs ~20s with thinking off). Send think:false on the /api/chat call; Ollama ignores it for non-thinking models (verified on qwen2.5:7b), so it's safe and unlocks the stronger Qwen3 family.
Code-audit clean-ups: share one normCurrency between the router and the templates, lift the duplicated nearest-day resolver into formatters.resolveDayId, drop two needless as-unknown-as casts at the fillBookingWideFields call sites, restore routeExtraction's doc comment, and give the broker template readable names. Plus recognise ¥/JPY and fall back to a standalone symbol amount, so a Klook-style voucher whose price sits far from any label still yields a cost.
reservation_endpoints.lat/lng are NOT NULL, so saving a reviewed transport whose pick-up/return couldn't be geocoded threw a 500 and lost the whole booking (dates, linked cost). Skip those rows; the dates still persist on reservation_time/reservation_end_time.
Apply the deterministic confirmation-code and total fill to vendor-template results too (not just model output), and require the captured reference to contain a digit so a bare 'Confirmation'/'Reference' label no longer grabs the next prose word.
Pull the hotel/rental fields these vendors print in a stable text layout (name, address, stay/pickup dates, price, reference) deterministically, so the import stops depending on the local model for them. Handles German long/abbreviated months and English dates incl. 12-hour and comma forms.
Parsing a booking can take a while on a CPU host, so don't hold the
upload modal open for it. The async import endpoint returns a job id
right away; the parse runs server-side (one at a time per user) and
pushes progress over the user's WebSocket, and a small widget in the
bottom corner tracks it while the user keeps navigating and editing.
A finished job opens the per-item review from the widget.
The single-shot prompt was unreliable on multi-leg flights and longer
documents, and slow on a CPU host. For the local provider, run a small
router instead:
- deterministic vendor templates first, with no model call at all
- exactly one grammar-enforced call per document via Ollama's native
`format` (flights as a flat array of legs, everything else as one flat
reservation, the type picked from keywords or a union schema)
- booking-wide fields (booking reference, total price, the overnight
arrival day) filled deterministically from the text afterwards, and
dates coerced to ISO so a natural-language date can't slip through
Recommend qwen2.5 in the AI-parsing settings instead of NuExtract.
NuExtract isn't an instruct model — fed a plain chat prompt it just echoes the
schema back. Detect a NuExtract model by id and talk to it the way the model
cards document: the JSON template inlined in a single user message, no system
prompt, no json_schema, temperature 0. Its flat result is mapped back to the
same KiReservation shape the rest of the pipeline already uses, so nothing
downstream changes; every other model keeps the generic prompt.
Money is taken as a verbatim string and parsed locally (German "1.580,22 €"
otherwise comes back as 1.49772), a rental car's pickup/return ride the from/to
fields so a stray form label doesn't become the location, and a lodging with no
name falls back to its address instead of being dropped.
When a confirmation carries a total price, record it as a real expense
linked to the reservation (in the matching Costs category) instead of
leaving the amount in metadata only. Gated on the Costs addon.
Request and map root-level seat/class/platform and a total price/currency into reservation metadata (shown on the card; price reuses the existing label). Read both the root and reservationFor and tolerate common field-name aliases (priceAmount, priceCurrencyISO4217Code, fareClass, ...) since models name these inconsistently. Also capture event/attraction venue telephone + url onto the auto-created place, matching lodging/restaurant.
On a GPU-less host the model's prompt-eval time scales with input length and dominates total latency. Booking details sit at the top of a confirmation, so capping the extracted text at 4000 chars (was 8000) roughly halves extraction time (~50s warm for a capable local 7B model) with no loss of fields on real hotel/rental confirmations. Tunable if a long multi-segment itinerary needs more.
- rental car: request+map dropoffLocation, emit pickup->return from/to endpoints, set a location string (G1/G2/G3). - geocode endpoints (stations/stops/terminals/rental desks) on confirm via Nominatim; mapper now emits coordless named endpoints and confirm persists only the geocoded ones (G6). - assign every dated booking to the nearest trip day so it still shows when slightly out of range, and keep hotel accommodation from vanishing when a check date misses (G5/G10). - fix bus mislabelled as train + add bus_number metadata (G7/G8), flag malformed boats (G9), accept root start/end time for events (G11). - raise the local-LLM timeout to 300s for CPU-only Ollama.
client: the import call inherited the global 8s axios timeout and aborted long LLM extractions even though the server finished it; remove the timeout. server: raise the OpenAI-compatible LLM timeout 60s->180s (a cold Ollama model can take ~45s to first token). server: cap extracted text to 8000 chars before the LLM - multi-page T&C tails (30k+ chars) overflowed the context window, truncating the relevant head and making CPU inference crawl; booking details sit at the top.
Booking import only fell back to the LLM when each user flipped an 'always retry with AI' toggle, so by default files kitinerary returned nothing for just failed. Run the fallback automatically whenever the AI Parsing addon is on (fallback-on-empty); drop the now-redundant per-user toggle and its setting.
Maurice