The proof of work and the bounded read of a 403 body recurse instead of looping, the background-work checks compare against null instead of testing a promise, and the tests repeat their requests through oneByOne. The renewal test loads the stored key before its first request, and the panel test picks the last confirm button without Array.prototype.at.
An expired key is no longer sent; the request goes through the shared pool
while a due key, or an expired one within its grace, is renewed in the
background. Refusals in one burst count once, a refused renewal counts like
a refused search, and a Places request whose key is refused is sent once
more without it.
docker-compose.yml and the Helm chart name the switch and the key (the key in the Secret, never the ConfigMap), and the wiki explains the installation key, what it carries, where it goes and the admin view of the allowance.
A new service-identity domain registers the installation with places.liketrek.com on its first Places request, in the background and behind a light scrypt proof of work, keeps the key encrypted in app_settings, renews it from day 60 through a daily job, drops it after three 401s in a row with up to six hours of jitter, and freezes it on a revocation until an admin starts over. TREK_SERVICES_KEY wins and turns registration off; TREK_SERVICES=off sends no key; a mirror set with TREK_PLACES_URL never registers and never sees the key, which goes only to the hosts in TREK_SERVICES_HOSTS over https.
The Places client takes its headers from the identity port, sends a TREK/<version> User-Agent and keeps the instance token. A 401 or 403 does not open its breaker. GET /api/admin/service-identity reports the state, tier and today's allowance, asking the issuer at most every 20 seconds; POST .../regenerate starts a new identity.
Without an Access Key the image search goes through Unsplash's web endpoint,
which refuses datacenter addresses with a 401 or 403 or answers with a
challenge page. The search then said only "Unsplash search unavailable",
although an admin can already enter a free key in the settings. Such a
refusal now answers 502 with that hint and the code unsplash_key_needed, and
the cover search shows it in the user's language.
Closes#1449
The candidate a GPX or KML row is checked as is built in lineCandidate, so
places.service.ts stays within its size baseline, and the atomic import test
imports the same Ridge line it later expects to be skipped.
Two GPX tracks a watch named alike (two "Afternoon Hike"s) were taken for one:
the import compared a track only by its name, so the second was skipped as
already in the trip. A track now matches a track of that name only when it
also starts, ends and reaches as far as that one, to about a hundred metres,
so the same file imported again is still skipped. The GPX, KML and tour
imports and the tour preview all read the same rule.
Closes#2870
The assistant sent an own server's chat to the address as entered plus
/chat/completions. Ollama, entered as http://host:11434 the way its own docs
write it, answers that with a 404, so every message ended in "something went
wrong" while connecting had worked: the model list is found under the root
at /api/tags. A bare server address now chats under /v1, where Ollama, LM
Studio, vLLM, llama.cpp and LiteLLM all answer; an address with a path, such
as https://api.groq.com/openai/v1, is used as entered.
A book held at most 150 spreads. The auto layout cut everything past that
without a word, so a long journey's book ended about three quarters of the
way through, with its last page and back cover gone, and adding, copying or
applying a designed page to a full book was refused.
The spread cap is gone from the book schema, the page operations, the
design composer and its apply, the assistant's book card and the design tool
contracts. A design's own page budget (pages.max) still bounds what it lays
out. The save route's 8 MB body still bounds a book's size, which at two to
five kilobytes an auto laid page is a few thousand pages.
The templated design digests move because their counter ids shift by two;
the pages, their order and their elements are unchanged.
With Google as the transit provider, the result list still credited
Transitous, and a route added from the panel or through
create_transit_journey was saved with provider 'transitous'. The panel
now credits "Google Maps" (untranslated, as Google's terms ask) or
Transitous after the backend that answered and saves that one; the MCP
journey records the backend that plans routes for the caller, and
search_transit_routes reports it.
Gemini 3 signs a tool call on its OpenAI-compatible endpoint
(extra_content.google.thought_signature) and refuses the next request
with a 400 when the call comes back without it, so every turn in which
the model called a tool failed. The Chat Completions client now keeps
the signature per call as the step's provider state and sends it back
with the call, to the model that signed it only.
An own server entered with Gemini's endpoint now also gets Gemini's
narrower tool schemas, which it refused before.
The Tours import takes over a same-named track a trip kept from 4.x, but a
route file attached in the assistant chat was read with the plain duplicate
rule only, so that track was reported as already in the trip and never
proposed. The read now leaves such a track to the import, which takes it
over as a tour like the Tours dialog does.
Refs #2810
remember_for_trip runs at once during a turn, and read_attached_file and
keep_attached_file are answered by the server, but all three are calls the
model made. Rebuilding a thread from its stored steps left them unanswered,
and ChatGPT, the OpenAI reasoning models and Anthropic replay their own record
of each step, call included, so the provider refused the conversation with a
400: the continuation after applied changes failed, and every later message of
the thread with it. A step that kept a note also sent an empty user message.
Every call the model made is now answered as in the live turn, the attached
file's outcome as the answer of its call; threads that broke this way work
again on their next message. A refusal's log line names the refused field's
path, never its message.
Closes#2845
create_day_note and update_day_note described `time` as a clock label and capped
it at 250 characters, while the planner shows it as the Markdown body under the
title and allows 2000. The assistant put long text in the title and had to cut
notes it was asked to edit. Both tools now take text and time from the REST
contract with descriptions naming title and body, and an update still refuses an
empty title. The assistant's change card labels the two fields title and content
instead of text and time.
Closes#2840
The airport picker never offered Little Cayman (LYB): OurAirports marks it as
having no scheduled service although it has daily flights, and the build
dropped every airport on that flag. The same flag also hid Vestmannaeyjar,
Western Sydney and around 1,280 other large and medium airports with an IATA
code. The build now applies the flag to small airports only, where it keeps
private airstrips out.
The bundled list gains exactly those airports. Every entry it already carried
stays unchanged, because a fresh OurAirports export currently renames PBI to
DJT and drops nine other codes that saved bookings point at.
Closes#2820
A GPX or KML import skipped every row whose name was already a place of the
trip. A watch names a hike after where it went, so "Hokitika Gorge" was
refused as already in the trip because the gorge itself was a planned
place, with no track to show for it (#2810). A track now matches only the
tracks of that name; a pin still matches every place of its name.
The Places filter also hid its "Tracks" choice whenever Tours was on, on the
grounds that tracks are tours then. A track imported before Tours is a
plain place and stays in the list, so the choice now counts the tracks that
are no tour.
A GPX track imported in 4.x is a plain place: a line, no tours row. It never
showed under Tours, and importing the same file there was refused as a
duplicate by its name, so there was no way to turn it into a tour (#2810).
The tour import now takes over a track of the trip that has the same name
and is no tour yet: the place keeps its name and line and gets the tours
row, with the recorded times the file adds. The preview says so instead of
"Already in the trip". New and taken-over tracks commit in one transaction.
Connecting an own server asked Ollama's /api/tags and gave up on any answer
that was not a 404, and every failure, a refused key included, read as "not
responding". The model list is now looked for at /api/tags, at the address as
entered and at /v1/models, an answer only counts when it is a list, and a
refused key or a server that answered without a list say so (#2808).
AI parsing through an own server always called Ollama's /api/chat, which an
OpenAI-compatible server such as Groq answers with 404. A server without it
now gets the same extraction over /chat/completions at the address as
entered (#2806).
TREK 5.0 left most registry plugins blocked because their authors still declare
"<5.0.0", and the only way past that was TREK_PLUGINS_IGNORE_TREK_RANGE for every
plugin at once. An admin can now install or switch on one such plugin after a
warning; the consent is kept on its row for that plugin version and TREK major,
so an update or a new major asks again. The environment switch works as before.
Discover marks a plugin not confirmed for this TREK on its card, and Install opens
the warning first. Admin > Plugins reads cleaner: one row of alike facts per plugin
with each state that needs attention on its own line, type, status and order behind
one Filter button, and the containment notes in a dialog behind the shield in the
header. The row, the filter menu and the dialog moved into their own files.
PocketBase answers a back-relation with a single record as the object itself,
not a list, so the user lookup failed as an answer TREK did not understand and
every search with it.
Closes#2799
TREK's code for Brazilian Portuguese is 'br', which Intl reads as Breton.
The assistant passed it straight to Intl.DisplayNames for the system
prompt's answer language and to the card formatters, so the model was told
to answer in Breton. Convert TREK codes to their BCP-47 tag with
getIntlLanguage before handing them to Intl.
Closes#2800
A token whose account has no local ActivityPub author was reported as
invalid_response, which the settings show as 'The answer did not come from
wanderer', although wanderer had answered fine. It now has its own code,
no_author, with a message in every language that says what is missing.
A failed wanderer call kept its reason in WandererError.detail (the field that
did not match, or what came instead of JSON), and the client comment said it
stays in the log, but nothing wrote it anywhere. The log only had the generic
'wanderer answered with something TREK does not understand', so neither an
admin nor we could tell what wanderer had sent. Every refusal now logs its
code, message and detail.
gpt-5.6-sol, gpt-6.1-sol and the other reasoning models refuse function tools
on Chat Completions, so every assistant message on such a model came back as
HTTP 400. An OpenAI connection whose model is from the o-series or GPT-5 and
later now goes through the Responses API, the client a ChatGPT plan already
uses, with the key as its bearer token and no token refresh on a 401. Older
models and the -chat variants stay on Chat Completions.
Closes#2792
Sonnet 5.5 and Opus 5.5 think by default and write what they say between tool
calls into thinking blocks, which they want back unchanged with the rest of
the turn. The client dropped those blocks (and their signatures) and rebuilt
the turn from its text and calls, so after a tool result the next step came
back empty and the assistant gave up with empty_answer. A turn that thought
is now replayed block by block like one that searched the web; the cache mark
before the question skips a thinking block, and replayed blocks are copies so
a mark never sticks to the stored answer.
Closes#2789
The model picker in Admin > Addons did nothing for a ChatGPT connection: the
only way to change the instance's model was saving the connection again, which
needs an API key a plan does not have. A PATCH /api/admin/ai/connection/model
now switches it on its own, to a model the provider offers.
Closes#2783
The overview of a trip's tours decides for itself when it shows, which keeps the planner page hook inside its complexity and length limits, and adds nothing to the camera's key while there is no tour to frame. The elevation helper names its empty case for the strict type check, the self-host wording exemption for the retired Dawarich card goes, the day plan test reads the wrapping pills, and the new tests drop their non-null assertions and loops.
The update dialog now shows the 5.0.0 headliners, Tours, TREK Studio, the assistant and help on every screen in 27 languages, each with a drawing of its own, translated into every language TREK ships. The 4.3.3 cards and their drawings are gone.
Dismissals were compared with the prerelease part dropped, so a notice closed on 5.0.0-pre.2 counted as closed on 5.0.0 and the final release showed nothing to anyone who had tried a prerelease. Versions now compare in full.
Each member's amount was converted and rounded to a cent on its own, so four 34,000 ISK quarters of a 136,000 ISK bill came to 997.52 EUR against the 997.51 the payer was credited. The balances ended a few cents off zero and settle-up had no payment for them. A split whose amounts match the payers in the bill's own currency is now converted as one set by largest remainder, in the settlement and in the per-person summary alike.
Six upload routes refused the shared demo account in demo mode; journey photos, videos, gallery media and covers, collab note files and chat images did not, so any visitor could put images in front of every other one for the hour. They answer the same 403 now, checked in the handler after the upload is read, and a text-only chat message still goes through. The Demo Mode wiki page lists them, and says document sync is not covered.
Maurice