Release-cycle test audit: the /mcp auth's constant-time token guard, IP keying,
ACCESS-type pinning, and brute-force message coupling were untested. Extract
behavior-preserving pure helpers so they're testable and cover them:
- sharedTokenMatches: length-mismatch early-returns before timingSafeEqual
(which throws on unequal lengths); equal-length uses timingSafeEqual; array
header -> first element; non-string -> false.
- clientIp: req.ip > socket > first XFF hop > 'unknown' (limiter keying).
- bindAccessJwtVerifier: verifyJwt pinned to JwtType.ACCESS (rejects REFRESH).
- CREDENTIALS_MISMATCH_MESSAGE single source of truth shared by
verifyUserCredentials and isCredentialsFailure, so a reworded auth error can't
silently disable the /mcp brute-force counter.
- verifyUserCredentials no-side-effect contract asserted via a TS-AST spec
(AuthService can't load under jest): its body has no createSessionAndToken/
audit/updateLastLogin while login() has all three.
Extractions are behavior-preserving (reviewed); class delegates to the helpers,
dead code + unused imports removed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Release-cycle review found the /mcp Basic path skipped the controller's
pre-token gates and over-eagerly minted sessions:
- SSO/MFA bypass (blocker): the Basic path called AuthService.login/
verifyUserCredentials directly, but validateSsoEnforcement + the lazy EE MFA
gate live in AuthController.login. Now enforceBasicLoginGate runs in the Basic
branch BEFORE any token is minted: validateSsoEnforcement(workspace) (reject
on enforced SSO) and the same lazy-require MFA check the controller uses
(reject MFA users -> 'use a Bearer access token'). No EE module bundled (this
fork) -> no MFA gate, identical to the controller; a throw from the check
fails closed (no token). Bearer/service-account paths are not gated (those
JWTs are minted post-gate).
- Non-init session mint: isSessionInit is now (no mcp-session-id) AND the body
is a real JSON-RPC initialize (isInitializeRequestBody). A header-less
non-initialize request takes the side-effect-free verifyCredentials path -> no
user_sessions row, no USER_LOGIN audit, no lastLoginAt bump.
- FailedLoginLimiter.sweep() now runs on an unref'd 60s interval, cleared on
module destroy (was never scheduled -> unbounded Map growth under XFF rotation).
- Subsequent (non-init) valid login no longer resets the global per-email brute
bucket (only per-IP / per-IP+email); the email backstop is reset only on a
deliberate init login.
Note: in a hypothetical EE build, checkMfaRequirements is called with no
FastifyReply (we only read requirement flags); a res-dereferencing EE impl would
surface as a clean rejection (fail-closed), not a bypass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The embedded MCP server acted as a single service account; now each /mcp
session authenticates as the current user, so tools run under that user's
CASL and edits attribute to them.
- HTTP Basic (chosen path): Authorization: Basic email:password, validated
server-side via AuthService; the session carries the issued user JWT (not
the raw password). Password may contain ':' (split on first only).
- Bearer fallback: Authorization: Bearer <access JWT>, verified as ACCESS and
additionally checked for an active session + non-disabled user (matching
JwtStrategy), so revoked/disabled users are rejected.
- Service account stays as an optional fallback (no creds + env configured).
- packages/mcp createMcpHttpHandler accepts a per-request config resolver
(back-compat: static config / stdio unchanged); identity is bound to the
mcp-session-id at init and re-validated from the caller's own credentials on
every request (anti session-fixation: a guessed session id can't be reused
without matching creds).
- A full login (session + audit) happens only once at session init; later
requests re-verify credentials via a new non-side-effecting
AuthService.verifyUserCredentials (no session/audit spam).
- Failed-login limiter (5/60s, keyed per-IP, per-IP+email, and per-email so IP
rotation can't brute one account) since direct login bypasses the controller
throttler. Only real credential failures count.
- MCP_TOKEN shared guard moved off Authorization to an X-MCP-Token header
(timing-safe compare); credsConfigured 503 gate replaced by a clear 401.
- No secrets logged; all auth resolved before res.hijack() so failures return
clean 401 JSON. .env.example marks the service account optional.
Implements docs/backlog/mcp-per-user-auth.md (variant L).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
These import paths relied on the private EE module that was deleted from
the repo. In the community build they either threw 'enterprise license'
(DOCX/PDF) or silently no-op'd (Confluence). The frontend buttons were
already removed in 38064064; this cleans up the dead backend stubs.
- import.service.ts: drop processDocx/processPdf methods, their dispatcher
branches, the pageId computation + insertPage spread, and the now-unused
moduleRef param/ModuleRef import
- file-import-task.service.ts: drop the Confluence branch and the now-unused
moduleRef param/ModuleRef import
- import.controller.ts: restrict file extensions to .md/.html and zip
sources to generic/notion; update the error message accordingly
- file.utils.ts: remove Confluence from the FileImportSource enum
- features.ts: remove the unused CONFLUENCE_IMPORT/DOCX_IMPORT/PDF_IMPORT
feature keys
The isConfluenceImport logic in import-attachment.service.ts is intentionally
left in place (real shared attachment-parsing code, not a stub); its removal
is a separate, riskier refactor.
The two catch blocks in importPage() threw an opaque "Error processing file
content" / "Failed to create imported page" BadRequest, hiding the real cause
from the HTTP response. This made a production 400 regression impossible to
diagnose without server log access, and violated the project convention that
errors must never be swallowed.
Extract `${err.name}: ${err.message}` into both the log (full err object kept
for the stack) and the thrown BadRequestException. Inner processMarkdown/
processHTML rethrowing catches and the EE processDocx/processPdf license
catches are left unchanged.
Local reproduction of the happy-dom 14->20 theory failed (full import chain
+ 22 edge cases pass on happy-dom@20.8.9), so the root cause is still pending
the now-visible reason from a recurring 400. Diagnostic script test-import.tsx
added; backlog doc updated with findings.
- Remove automatic panel opening in handleAddComment
- Remove automatic panel opening in handleAddReadOnlyComment
- Keep panel open on click for existing comments in editor
Reduce DocTree row stride from 32px to 26px for a denser sidebar tree,
and fix the selection/hover highlight that looked unbalanced at the
tighter spacing.
Root causes:
- The virtualized <li> had no explicit height, so `.node`'s height:100%
collapsed to content height; combined with the asymmetric
`[role="treeitem"] { padding-bottom: 2px }` rule, row content was
pushed to the top of the highlight pill (icon glued to the top edge).
- NodeMenu / CreateNode action icons used the default Mantine ActionIcon
size (md = 28px), overflowing the tighter 26px row stride onto
neighbouring rows.
Changes:
- doc-tree.tsx: rowHeight 32 -> 26; give each row <li> a definite
height = rowHeight.
- tree.module.css: rowWrapper fills the slot (height:100%); node pill is
inset and vertically centered (height: calc(100% - 4px)); drop the
asymmetric [role="treeitem"] padding-bottom.
- space-tree-node-menu.tsx / space-tree-row.tsx: action icons size={20}.
- share.module.css: drop now-dead .treeNode padding-bottom override.
Verified in an isolated browser harness: highlight content is centered
(2.8/2.8px) and nothing overflows the row stride.
Reduce the recent/favorites/created-by-me list tables from
verticalSpacing="sm" (12px Td padding) to a numeric 6px, removing
~12px of extra height per row so the home page lists pack closer
together. The shared RecentChanges table also drives the space home
view, so both stay consistent.
Drop the Overview home link that sat between the space switcher and
the Pages section in the authenticated space sidebar. Remove the JSX
block and clean up the now-unused imports (UnstyledButton, IconHome,
useLocation, getSpaceUrl) and the local `location` variable.
Reduce DocTree default rowHeight from 32px to 26px so sidebar page
tree rows pack closer together. The virtualizer uses rowHeight as the
row stride (estimateSize + translateY), and row content is only ~22px
tall, so the previous 32px left an ~8px gap between nodes. Both the
space tree and the shared (public) tree inherit the default, so both
become denser and stay consistent.
Rework the space sidebar:
- remove the "New page" and "Search" menu items (search stays in the app
header; page creation stays via the "+" button in the Pages section)
- move "Space settings" into a gear icon next to the current space name
- drop the searchable space popover and render all spaces as an
always-visible grid of fixed-width cards (icon + name), several per row,
sorted alphabetically, with the active space highlighted
- always inject the active space into the grid so it stays highlightable
even when the user has more than the 100-space API page limit
The shared SpaceSelect component is left untouched (still used by the
move/copy page modals).
list_pages gains an opt-in `tree` parameter on both surfaces (the
@docmost/mcp server tool and the AI-chat agent tool), which share the
same DocmostClient.listPages. Default behavior (recent-by-updatedAt flat
list) is unchanged.
- client.ts: listPages(spaceId?, limit=50, tree=false); when tree is
true it requires spaceId (throws a specific error otherwise), walks the
sidebar tree via the existing bounded/cycle-safe enumerateSpacePages,
and returns a nested tree; limit is ignored in tree mode.
- lib/tree.ts: new pure buildPageTree() — lean nodes { id, slugId, title,
children? }, children sorted by position (code-unit order), orphans
promoted to roots, cycle-safe.
- index.ts + ai-chat-tools.service.ts: expose `tree` in the tool schemas
and descriptions; docmost-client.loader.ts: mirror the new signature.
- tests: add packages/mcp/test/unit/tree.test.mjs (nesting, ordering,
lean shape, orphan promotion, cycle/self-reference safety).
- rebuild @docmost/mcp (build/ is tracked and loaded at runtime).
Wrap the logo link and the APP_VERSION text into a single bottom-aligned
Group so they read as one lockup ("gitmost v0.9..."). Move the version
styling into a new .brandVersion CSS class: shrink it from 12px to 10px,
keep the dimmed color and selectability, and lift it via margin-bottom so
its text baseline sits on the wordmark baseline of the 30px desktop logo
(derived from the logo SVG geometry). Drop the redundant lh prop.
The transcript force-scrolled to the bottom on every streamed delta because
the auto-scroll effect ran unconditionally whenever the messages array identity
changed. Scrolling up to read earlier messages was impossible — each token
yanked the view back down.
Implement a "stick to bottom" pattern in MessageList:
- track whether the viewport is pinned to the bottom via a scroll listener
(pinnedToBottomRef, BOTTOM_THRESHOLD = 40px);
- only auto-scroll while pinned; a freshly sent user message always re-pins;
- attach the scroll listener via a [hasScrollArea] dependency so a brand-new
empty chat (whose ScrollArea mounts only after the first message) wires it up;
- guard the effect's own scrollTop write (programmaticScrollRef) so it is not
misread as a user scroll.
The floating AI-chat header badge summed metadata.usage (AI SDK
totalUsage, all steps) across every assistant row, showing the
cumulative tokens SPENT — which grows each turn as history is re-sent.
Replace it with the conversation's CURRENT context size.
- server: persist metadata.contextTokens in streamText onFinish from the
final-step `usage` (inputTokens + outputTokens ≈ current context
window occupancy); keep usage: totalUsage for back-compat/fallback
- client: derive the badge from the most recent assistant row's
contextTokens (fallback to that row's usage total for older chats)
instead of summing all rows
- types: add metadata.contextTokens to IAiChatMessageRow
- i18n: rename badge label "Tokens used in this chat" -> "Current
context size" (en-US)
No DB migration needed (metadata is a JSON column).
Replace the implicit `hostname endsWith openrouter.ai` detection with an
explicit, admin-chosen provider field `sttApiStyle` ('multipart' = OpenAI-
compatible multipart /audio/transcriptions; 'json' = OpenRouter-style JSON +
base64 input_audio). The transcription path now branches on the stored field,
not on the URL — nothing hidden from the admin.
- ai.types: add SttApiStyle + STT_API_STYLES; field on AiProviderSettings and
MaskedAiSettings (resolved via ResolvedAiConfig).
- update-ai-settings.dto: validate sttApiStyle with @IsIn(STT_API_STYLES).
- ai-settings.service: plumb sttApiStyle through resolve()/getMasked() and the
non-secret update whitelist; workspace.repo: add it to the ALLOWED array so it
persists.
- ai.service: drop isOpenRouter(); transcribe() branches on cfg.sttApiStyle;
rename helper to transcribeJsonBase64 with provider-neutral error text and a
BadRequestException (400) when the base URL is missing for the JSON style.
- client: SttApiStyle type on IAiSettings/IAiSettingsUpdate; "Request format"
Select on the Voice/STT settings card; i18n.
- ai.service: route *.openrouter.ai STT to its JSON+base64
/audio/transcriptions API; keep the OpenAI multipart path (AI SDK) for
OpenAI/self-hosted whisper. Unify transcription behind transcribe().
- /transcribe controller: surface the real provider/transport reason
(describeProviderError) instead of an opaque 500; preserve HttpException.
- testConnection: add an 'stt' capability (silent-WAV probe) + DTO; client
gets a Test endpoint button and status dot on the Voice/STT card.
- useDictation: log full errors to the console and show the real reason
(mic start + transcription paths); handle NotReadable/Abort and missing
mediaDevices.
- docs(CLAUDE.md): require full error logging + specific user-facing messages.
Add push-to-talk voice dictation that transcribes recorded audio on the
server via the workspace's OpenAI-compatible AI provider (Whisper /
gpt-4o-transcribe / self-hosted whisper), then inserts the text.
Backend:
- New `stt_api_key_enc` column + migration; STT creds parity with chat/
embeddings (sttModel/sttBaseUrl/sttApiKey, write-only key, fallbacks to
chat baseUrl/key). Both provider whitelists updated (service + repo).
- AiService.getTranscriptionModel + AiTranscriptionService.
- Gated POST /ai-chat/transcribe (dictation flag → 403, JWT + workspace
scope + throttle, 25MB cap, MIME whitelist, never logs audio/key).
- New `settings.ai.dictation` workspace flag (DTO + service + audit).
Frontend:
- Wire up the Voice/STT settings card (model/base URL/key) and the
Voice-dictation toggle.
- New `features/dictation`: useDictation (MediaRecorder state machine),
MicButton, transcribe service; integrated into the chat composer and a
new editor-toolbar dictation group, both gated by ai.dictation.
Make the floating AI chat window open at a larger default size and
allow stretching it further, plus shrink the fonts.
- ai-chat-window.tsx: DEFAULT_WIDTH 362->540, DEFAULT_HEIGHT 602->680;
clamp the default width to the viewport in computeInitialGeom()
(symmetric with the existing height clamp) to avoid overflow on
narrow screens.
- ai-chat-window.module.css: raise resize caps (max-width 560->900px,
max-height 880->1100px); base font-size 12->11px.
- ai-chat.module.css: chat content font .messages sm->xs.
Add a header button to the AI agent chat window that copies the active
conversation to the clipboard as Markdown, including the request
internals already persisted client-side — tool calls with their
input/output, per-message token usage, and finish/error info. No new
network call and no server/DB change: it serializes the already-loaded
persisted message rows.
- New util chat-markdown.ts (renamed from export-chat.ts): pure
buildChatMarkdown() serializer reusing the tool-parts helpers so tool
labels match the on-screen labels; fence() escapes embedded code
fences.
- ai-chat-window.tsx: Copy button (shown only for a saved chat with
loaded rows) using the project useClipboard hook; toggles a check
icon on success and shows the standard "Copied" notification. Drag is
unaffected (startDrag ignores button clicks).
- en-US: add "Copy chat" key, drop the obsolete "Export chat".
Move the AI-chat toggle icon (IconSparkles) from the page header menu
into the global top bar, placed next to the notifications icon. The
"AI chat enabled" gate (workspace.settings.ai.chat) is preserved, and
the icon style is aligned with the neighbouring notifications icon
(subtle, size sm). As a result the entry point is now available on all
routes instead of only on page routes.
- app-header.tsx: render the gated AI-chat ActionIcon before
NotificationPopover; wire it to aiChatWindowOpenAtom.
- page-header-menu.tsx: remove the old AI icon block and its now-unused
imports/locals.
Add autoFocus to the chat composer Textarea so a freshly created chat
(window open, "New chat", chat switch — all remount ChatThread via key)
lands with the cursor ready in the input field, letting the user type
immediately without clicking into it.
edit_page_text reported "success" when asked to change formatting (e.g. remove
strikethrough): the markdown-strip fallback matched the bare text, the replace
preserved marks, and the tool returned success — so the agent believed it had
fixed something that never changed.
Two fixes, both in the shared @docmost/mcp DocmostClient so they reach BOTH the
standalone MCP server and the in-app AI chat (which loads @docmost/mcp):
- Verifiable result for every content mutator: mutatePageContent now computes a
`verify` change-report (text inserted/deleted, blocks changed, per-mark-type
delta, integrity/structure delta) via summarizeChange() and returns it on all
mutators (incl. replaceImage via mutateLiveContentUnlocked). diffDocs is
text-only, so the mark/structure delta is what surfaces formatting changes.
- edit_page_text hard-refuses formatting edits: applyTextEdits rejects an edit
whose find/replace differ only in markdown markers (via stripBalancedWrappers,
which strips balanced wrappers/links without trimming whitespace/emoji, so
plain-text edits like trailing-space trims, snake_case, math are NOT refused).
A fully-refused batch errors instead of silently succeeding.
Also updated the model-facing edit_page_text descriptions in BOTH tool layers
(packages/mcp/src/index.ts and ai-chat-tools.service.ts) to drop the misleading
"strip-and-retry tolerated" wording and point formatting changes to patch_node.
New unit tests: test/unit/diff-verify.test.mjs, test/unit/json-edit-refuse.test.mjs.
The sidebar page tree previously kept its expanded/collapsed state in an
in-memory jotai atom, so it reset on every reload. Persist it in
localStorage instead, scoped by `${workspaceId}:${userId}` so multiple
accounts sharing a browser origin don't leak tree state into each other.
- open-tree-nodes-atom: replace the in-memory atom with a facade over an
atomFamily of atomWithStorage, scoped via currentUserAtom; keep the same
[OpenMap, functional-updater setter] public API so space-tree needs no
change.
- Use an explicit synchronous createJSONStorage + getOnInit: true so the
saved state is read at atom init — no collapse-then-expand flicker on
reload and no write against an un-hydrated empty map.
- README / README.ru: document persistent page-tree state in the
"What's different from Docmost" table.
Add an "Export chat" button to the AI agent chat window header that
downloads the active conversation as a Markdown file. The export is
client-only: it serializes the already-loaded persisted message rows
(no new network call, no server/DB change) and includes the request
internals the chat already holds — tool calls with their input/output,
per-message token usage, finish reason and error info.
- New util apps/client/src/features/ai-chat/utils/export-chat.ts:
buildChatMarkdown() + exportChatAsMarkdown(); reuses tool-parts
helpers so tool labels match the on-screen labels; fence() escapes
embedded code fences; slugify() yields a safe filename with a chatId
fallback; downloads via file-saver's saveAs.
- ai-chat-window.tsx: IconFileExport button in the header, shown only
for a saved chat with loaded rows (canExport); drag is unaffected.
- en-US: add "Export chat" and "You" i18n keys.
Display the app version (output of `git describe --tags`) in the header
beside the gitmost logo: a clean tag renders as `vX.Y.Z`, otherwise the
tag plus commits-since and short hash (e.g. v0.90.1-56-g25975acd).
- vite.config.ts: resolve APP_VERSION from env (Docker/CI) -> git describe
(local) -> package.json version fallback
- app-header.tsx: render APP_VERSION after the brand block (ml="md"),
nudge the Home nav group (ml={50} -> "xl")
- Dockerfile: accept APP_VERSION build-arg in the builder stage (.git is
excluded from the build context)
- CI: pass APP_VERSION build-arg — release.yml uses the tag, develop.yml
computes git describe with fetch-depth: 0
- nx.json: add APP_VERSION to the build target inputs so the cache
invalidates when the version changes
Extract the AI provider/endpoints settings and the MCP server section out
of the Workspace "General" settings page into their own "AI" settings page,
reachable from a new sidebar entry.
- add page apps/client/.../settings/workspace/ai-settings.tsx (AiProviderSettings
admin-gated + McpSettings), with its own Helmet title
- register the /settings/ai route in App.tsx and add SETTINGS.WORKSPACE.AI
to app-route.ts
- add an "AI" item (IconSparkles) to the Workspace group in settings-sidebar
- trim workspace-settings.tsx back to the General section and drop the
now-unused imports
Rebuild the workspace AI settings page into card-based "Endpoints"
(Chat / Embeddings / Voice) matching the new design, and split the
single connection test into independent per-endpoint Test buttons.
- server: testConnection(workspaceId, capability) probes only the
requested capability ('chat' | 'embeddings'); add TestAiConnectionDto
and wire it through the /workspace/ai-settings/test controller
- client: testAiConnection(capability) + capability-typed mutation; two
independent test mutation instances so Chat/Embeddings results are isolated
- client: full rewrite of ai-provider-settings into Endpoints section —
drop the provider dropdown (driver is always openai, base URL + key
always shown), move the "AI chat" and surface the "Semantic search"
feature toggles into card headers, system message behind an Edit modal,
pgvector/reindex footer, and a disabled Voice/STT stub
- client: restyle external MCP tools and the MCP server section; collapse
the AI sections in workspace-settings; remove the standalone
ai-chat-settings component
- toggles now surface the server error message (e.g. missing pgvector)
- i18n: add new English strings
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
DOCX, PDF and Confluence import relied on a private EE module that was
dropped from this build, so those code paths only threw "enterprise
license" errors (DOCX/PDF) or silently did nothing (Confluence) while the
UI still presented them as working options.
- page-import-modal: drop the Word (DOCX), PDF and Confluence FileButtons
- remove the now-dead icon imports (IconFileTypeDocx, IconFileTypePdf,
ConfluenceIcon), the docx/pdf/confluence file refs and their input-reset
branches in handleFileUpload/handleZipUpload
- delete the orphaned confluence-icon.tsx component (no remaining importers)
Markdown, HTML, Notion and the generic zip upload remain unchanged.
Improve agent RAG quality with three changes, plus a roadmap doc for the rest.
- Indexer: prefix each chunk with its heading path ("Page > H1 > H2"), built by
walking the ProseMirror JSON (heading nodes) so a `#` inside a fenced code block
is never mistaken for a heading. Falls back to plain-text chunking on any error.
buildChunkRows: drop indexOf-against-source offsets (breadcrumb prefixes break
verbatim matching) for a cumulative cursor — offsets are provenance-only.
- Hybrid search: new migration adds a generated `fts` tsvector column + GIN index
to page_embeddings (same english+f_unaccent config as pages.tsv). New
PageEmbeddingRepo.hybridSearch fuses cosine + full-text rankings via Reciprocal
Rank Fusion (k=60, equal weights) in one SQL query at chunk granularity.
- Tools: collapse semanticSearch + searchPages into one hybrid `searchPages` tool
with a query-rewrite-oriented description; gracefully falls back to the REST
full-text path when embeddings are unconfigured. Access control (space scope +
page-permission post-filter) preserved. Add a query-rewrite hint to the default
system prompt.
- docs/rag-improvements-plan.md: record what shipped and the deferred backlog
(reranker, attachment indexing, eval harness, tuning).
Note: requires a corpus reindex to populate breadcrumbs on existing pages.
"Indexed N of M pages" stayed at e.g. "27 of 34" forever even after a
successful full reindex. The numerator counted pages that have embeddings
while the denominator counted ALL non-deleted pages, so empty / text-less
pages (which legitimately store zero embeddings) could never be reached.
Add PageRepo.countEmbeddablePages: counts non-deleted pages that have
non-empty textContent OR already have a stored embedding row, and use it as
the totalPages denominator in AiSettingsService.getMasked. The "has
embeddings" clause covers pages indexed from the content JSON (null
textContent) and guarantees indexedPages <= totalPages. No DB migration.
The floating AI chat window used a hard-coded border-radius of 14px,
larger than any other element and out of line with the rest of the UI.
Switch to the Mantine md radius token (8px) so the window corners blend
with the inner cards and message bubbles.
The bulk embedding reindex could hang on a single page forever
("Indexed 27 of 34 pages") with zero log output:
- all progress logs were debug-level, suppressed in production (pino info);
- embedMany() had no timeout, so a slow/hung embeddings endpoint blocked
the sequential per-page loop indefinitely.
Changes:
- ai.service.embedTexts: bound embedMany with AbortSignal.timeout
(configurable via AI_EMBEDDING_TIMEOUT_MS, default 120000ms); on timeout
throw a clear, greppable message, classified by both signal.aborted and
the error name (TimeoutError/AbortError/ResponseAborted) so a real
provider error racing the timer keeps its diagnostics.
- embedding-indexer.reindexWorkspace: promote lifecycle/progress logs to
info; log "[i/N] indexing page <id>" BEFORE the await so a hang names the
stuck page; warn on slow pages (>30s); add timing + final summary.
- .env.example: document AI_EMBEDDING_TIMEOUT_MS.
A misconfigured embeddings endpoint failed the RAG indexer with an opaque
"Invalid JSON response" and was not caught by "Test connection" (which only
probed the chat model), so it only surfaced silently during background
indexing.
- add describeProviderError(): formats AI SDK errors as
"<statusCode>: <message> | response body: <truncated one-line snippet>"
(statusCode/message/responseBody never carry the API key)
- use it in the bulk-reindex catch and the embedding processor's formatter so
the real cause (e.g. an HTML 404 from a wrong base URL) is visible in logs
- testConnection now probes chat AND embeddings independently: skips a probe
when that capability is unconfigured, returns ok:false with a Chat:/Embeddings:
prefix on real failure, "not configured" when neither is set
The WORKSPACE_CREATE_EMBEDDINGS / WORKSPACE_DELETE_EMBEDDINGS jobs were
enqueued (on AI Search enable/disable) but had no AI_QUEUE handler, so
existing pages were never indexed ("Indexed 0 of N pages") and disabling
never purged embeddings.
- EmbeddingProcessor: handle WORKSPACE_CREATE_EMBEDDINGS (bulk reindex all
live pages) and WORKSPACE_DELETE_EMBEDDINGS (purge workspace embeddings)
- EmbeddingIndexerService: add reindexWorkspace() (skips when embeddings
unconfigured; per-page error isolation) and removeWorkspace()
- PageRepo.getIdsByWorkspace(), PageEmbeddingRepo.deleteByWorkspace()
- AiSettingsService.reindex() + admin-only POST /workspace/ai-settings/reindex
- Frontend: "Reindex now" button, service call and mutation
- Stable per-workspace jobId with remove-before-add so a stale job can't
block future reindexes; cancel the delayed purge on enable/reindex so it
can't wipe freshly-built embeddings
Per-workspace AI provider config previously shared a single base URL and
a single API key between the chat model and the embedding model. Add
dedicated, optional embedding endpoint/token that fall back to the chat
values when empty, preserving backward compatibility.
- db: new migration adds nullable `embedding_api_key_enc` to
`ai_provider_credentials`; chat key stays in `api_key_enc`
- repo: add `upsertEmbeddingKey` / `clearEmbeddingKey` (on-conflict
touches only its own column, so chat/embedding keys never overwrite)
- ai-settings.service: store non-secret `embeddingBaseUrl`; resolve()
applies fallback (embeddingBaseUrl || baseUrl; embedding key || chat
key); getMasked() exposes raw `embeddingBaseUrl` + `hasEmbeddingApiKey`,
never the key; update() handles the embedding key write-only
- ai.service: getEmbeddingModel() builds openai/gemini/ollama with the
embedding-specific URL/key; chat path unchanged
- client: new "Embedding base URL" and "Embedding API key" fields with
fallback hints and a clear-key action
Requires running the DB migration on deploy.
The EE-frontend removal (a88b3f77) deleted enable-ai-chat.tsx, the only UI
that set the workspace flag settings.ai.chat. The backend gates remained:
the header assistant icon (page-header-menu) and the /ai-chat/stream endpoint
both require settings.ai.chat === true, and the flag is auto-enabled only in
cloud mode. As a result, self-hosted community builds could configure an AI
provider but had no way to turn the chat on, so the assistant icon never
appeared.
Restore a single admin toggle wired to the already-existing backend `aiChat`
field, modeled on the adjacent MCP settings switch:
- add aiChat?: boolean to the client IWorkspace type
- add AiChatSettings component (optimistic update, revert-on-error,
workspaceAtom sync so the icon updates without reload)
- render it (admin-only) in workspace settings between AI / Models and
AI / External tools (MCP)
Generative AI and AI search toggles, removed in the same commit, are left
out of scope.
Replace the bare brand text on pages with the Gitmost logo lockup
(mark + "gitmost" wordmark) and use the mark as the favicon.
- add generated logo lockups (text outlined from Space Grotesk SemiBold)
in dark/light ink variants; add reusable theme-aware <BrandLogo> component
- use BrandLogo in the global header (mark-only on mobile, full lockup on
desktop) and on auth pages, dropping the old Docmost icon + plain text
- point favicon to /brand/gitmost-favicon.svg (SVG primary + PNG fallbacks);
regenerate favicon/app-icon PNGs from the brand SVGs
- rename app name Docmost -> Gitmost (getAppName, index.html title/apple
title, manifest name); use getAppName() in the 404 title
- align theme/background colors to the brand tile (#0E1117)
- move brand guide and logos into docs/brand/ (canonical home) with a README,
and serve runtime copies from apps/client/public/brand/
Read tools (getPage, getPageJson, getNode, diffPageVersions,
exportPageMarkdown) return whole pages with no size cap. Their outputs
were stored verbatim in metadata.parts and the tool_calls column, and
metadata.parts is replayed to the provider on every later turn via
convertToModelMessages. After reading a couple of large pages the prompt
grew by full page bodies each turn — rising token cost, latency and DB
row size.
Add compactToolOutput(): a pure, recursive, size-bounded compactor used
in assistantParts() and serializeSteps(). It preserves the value's kind
and small scalar fields (id/title/pageId, which the client reads to build
citations on reload) while truncating long strings, capping long arrays
with a marker, and collapsing subtrees past a depth limit. Small outputs
are returned unchanged by identity. Tool inputs are left intact so
replayed tool_use arguments keep their object shape.
Compaction runs only at persistence time (onFinish/onAbort), so the live
stream and the current turn's multi-step reasoning still see full bodies.
Add unit tests for compactToolOutput.
Typing into the composer while the agent was streaming lost the draft once
the turn finished: on a brand-new chat, adopting the freshly created chat id
changes ChatThread's key and remounts it, wiping ChatInput's local state.
Lift the composer draft into a module-level jotai atom (aiChatDraftAtom) so it
survives the remount. Reset it only on deliberate chat switches — startNewChat,
selectChat, and the page-history "AI agent" badge deep-link — so a draft never
leaks between conversations, while adoption (which goes through a useEffect)
preserves it.
Display "Indexed N of M pages" on the AI provider settings page so admins
can see how much of the wiki is covered by vector-RAG semantic search.
- page-embedding.repo: add countIndexedPages() — distinct non-deleted pages
that have stored embeddings in the workspace
- page.repo: add countByWorkspace() — total non-deleted pages
- ai-settings.service: compute both counts in getMasked() (Promise.all) and
return them with the masked settings; inject PageEmbeddingRepo + PageRepo
- MaskedAiSettings / IAiSettings: add indexedPages + totalPages
- ai-provider-settings: render a dimmed coverage line under "Embedding model"
- i18n: add the "Indexed {{indexed}} of {{total}} pages" key (en-US, ru-RU)
The default system prompt introduced the assistant as embedded in
"Docmost". Update DEFAULT_PROMPT to say "Gitmost" so the agent
presents itself with the correct product name.
Surfacing the stream error via useChat().error alone was not enough: on a
brand-new chat the errored turn still fires onFinish -> onTurnFinished, which
adopts the freshly-created chat id and changes the <ChatThread> key, remounting
it with a fresh useChat whose transient `error` is gone. The thread re-seeds
from persisted history, where the assistant row has empty parts and the error
lives only in metadata.error — which was never rendered. Result: an empty
"AI agent" row and no visible error.
- Render the persisted metadata.error inline in MessageItem, so the error
survives the remount and is also shown in reopened chat history.
- Carry metadata.error onto the rebuilt UIMessage in rowToUiMessage.
- Extract the error formatter into utils/error-message.ts (describeChatError)
and reuse it for both the live Alert and the persisted error.
- Add metadata.error to the IAiChatMessageRow type.
Client-only; the server already persists metadata.error. No new i18n keys.
The AI chat UI previously collapsed every non-403/503 failure into a
generic "could not respond" message, hiding real provider errors such as
OpenRouter HTTP 402 "requires more credits". The backend already forwards
the real "<status>: <message>" via pipeUIMessageStreamToResponse onError,
so the fix is client-side.
- describeError now returns the provider message verbatim for any error
that is not one of our own gating responses, so 402 (credits), 429
(rate limit) and similar causes are visible to the user.
- Match gating responses by the NestJS JSON "statusCode" field instead of
loose substring/word checks, so a provider message that merely contains
"403"/"503"/"disabled" is no longer misclassified and hidden.
- Add a providerDetail() helper that filters empty text and the opaque
"An error occurred." / "Internal server error" placeholders, falling
back to the generic message only then.
No backend changes; no new i18n keys.
BullMQ rejects custom job IDs containing ':' (Redis key separator),
throwing "Custom Id cannot contain :" inside the onStoreDocument hook
for every agent edit. This broke agent-driven page saves (MCP
create_page runs as actor='agent') with HTTP 400.
Switch the agent dedup suffix from `${page.id}:agent` to
`${page.id}-agent`. The jobId is only used as a BullMQ dedup key and is
never parsed by the history processor; page.id is a UUID, so the
hyphenated id cannot collide with a human job whose id is a bare page.id.
Replace the docked right-aside AI chat with a floating, draggable,
resizable, minimizable window per the GitmostAgent design. The "AI chat"
entry points (page header menu, page-history item) now open the window
instead of the aside tab.
- Add ai-chat-window.tsx + ai-chat-window.module.css: fixed-position
window with viewport-clamped drag, CSS resize, minimize (hides body
via CSS so ChatThread/useChat stays mounted and streaming is not
aborted), and geometry kept in state (survives close/reopen, re-clamped
on open via useLayoutEffect, size tracked with a ResizeObserver).
- Reuse ChatThread, ConversationList and the transcript components
unchanged; move all orchestration (active chat, adopt-new-chat,
openPage, queries) into the window.
- Header shows a tokens-only badge: sum of persisted metadata.usage for
the active chat (no cost/context-total data available), hidden at 0.
- Add aiChatWindowOpenAtom; mount the window once in global-app-shell.
- Remove the aside "ai-chat" tab handling and delete ai-chat-panel.tsx.
- Type IAiChatMessageRow.metadata.usage; add "Minimize" and
"Tokens used in this chat" i18n keys.
Locators (edit_page_text `find`, insert_node `anchorText`) are matched
against the document's plain text, so a model-supplied locator carrying
markdown wrappers (**bold**, *italic*, `code`, [t](url)) or trailing emoji
never matched and the edit/insert failed. Add stripInlineMarkdown() and a
fallback: try the locator verbatim first (exact match wins, so literal
asterisks/underscores still work), and only on zero matches retry with a
markdown-stripped form. The ambiguity guard runs on the post-fallback count,
and `replace` / inserted node content are never stripped, so no formatting is
lost. Failed edits gain an atom-aware reason plus a bounded "closest block
text" hint; the insert_node "anchor not found" error now points at plain-text
anchors / anchorNodeId.
New packages/mcp/src/lib/text-normalize.ts (+ unit tests); wired into
json-edit.ts and node-ops.ts; tool descriptions updated. Tests: 212 pass.
edit_page_text (applyTextEdits) now matches at the inline-block level instead of
per text node, so a find/replace may cross bold/italic/link boundaries; the
replacement inherits marks from the unchanged common prefix/suffix via a diff
splice. Atom (non-text inline) slots can never be part of a match, making the
U+FFFC placeholder collision-safe, and inserted text never inherits an atom's
marks.
The edit batch is no longer all-or-nothing: applyTextEdits returns
{ doc, results, failed } and applies what it can; editPageText writes only on a
real change (no spurious history version for a no-op) and throws an aggregated,
actionable error only when nothing applied.
The AI-chat insert_node / patch_node / update_page_json tools now JSON.parse a
node/content argument that arrives as a string, matching the standalone MCP
server (this is what made insert_node fail under OpenAI tool calls).
Tool descriptions gain concrete ProseMirror examples and reflect the new
edit_page_text behavior. Adds/updates json-edit unit tests (183 pass).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
