Repository navigation
What's Changed
- Trace dependencies in Langfuse and SigNoz and prevent graph purge deadlocks by @justsml in #25
- Unify search into one surface; drop searchMany and /v1/search/batch by @justsml in #26
- Move agent-safety annotations into the capability manifest by @justsml in #31
- Add SDK coverage for runs, documents and streamed chat by @justsml in #32
- Generate MCP tool schemas from their zod sources by @justsml in #29
- Standardize the native error envelope by @justsml in #33
- Emit the error envelope directly from native handlers; document ApiError in OpenAPI by @justsml in #34
- Add opt-in LanceDB refinement with scoped service tests by @justsml in #35
- Separate retrieval candidate depth from reranking and prompt order by @justsml in #36
- Add opt-in metered DeepInfra reranking to workspace search by @justsml in #37
- Root each background loop pass in a named span by @justsml in #38
- Reduce Free starter credits to 250 by @justsml in #39
- Retire automation tools and collapse the MCP catalog onto public names by @justsml in #40
- Speed up CI with parallel test lanes on one Node LTS by @justsml in #41
- Drop the retired automation tables (contract step for #40) by @justsml in #42
- Improve retrieval evidence selection and bounded multi-search by @justsml in #16
- Harden MCP OAuth and add MCP SDK, Pi and openai client tests by @justsml in #44
- Specific errors for direct upload files/complete 400s by @justsml in #47
- Add the Doc Tumbler logo and brand kit by @justsml in #48
- Keep browsers on the consent page when MCP authorization fails by @justsml in #51
- Expect expires_at in the consent route test by @justsml in #53
- Accept real AWS access key IDs in production storage check by @justsml in #46
- Multipart uploads: equal-length non-trailing parts (R2 InvalidPart) by @justsml in #52
- Create pgcrypto in a new migration (0061) by @justsml in #50
- Env-tunable S3 client and separate LanceDB storage URI by @justsml in #43
- Trace LanceDB native S3 requests (LANCEDB_S3_TRACE) by @justsml in #45
- LanceDB on Backblaze B2 via s3+ddb:// (DynamoDB commit store, split credentials) by @justsml in #49
- Mount /etc/alternatives in the parser sandbox so ffmpeg can load by @justsml in #55
- Put the verification code in the email subject by @justsml in #54
- Refit the brand mark in the product palette by @justsml in #57
- Read silent videos from frames; name the failure when video is off by @justsml in #56
- Add pluggable parser sandbox providers (bubblewrap, landlock, modal) by @justsml in #62
- Strict chat model gating, per-request agent limits and GenAI agent tracing by @justsml in #65
- Add Nemotron 3.5 Super VL as an opt-in hosted vision route by @justsml in #64
- Drop legacy chat path from search-api.md by @justsml in #68
- Keep interrupted document stages retryable; stop flagging HeadObject 404s as errors by @justsml in #69
- Search: cache query embeddings and version-keyed vector results by @justsml in #70
- Measure which LanceDB S3 reads make up search latency by @justsml in #71
- Add document tags and consistent retrieval filtering by @justsml in #66
- Upgrade stripe to 23 in server by @justsml in #75
- Flatten MCP tool schemas so Claude chat models work; report multi-page evidence by @justsml in #72
- Read search availability chunk/page totals from trigger-maintained document counters by @justsml in #73
- Cut CI runner minutes (port of archived #196) by @justsml in #76
- Deferred indexing modes: page map and page images (core) by @justsml in #78
- Document versions, retention and config-change jobs (core) by @justsml in #79
- LanceDB: sized shared Session, opt-in prewarm, configurable read consistency by @justsml in #77
- LanceDB: read-through disk cache for immutable Lance files in the S3 proxy by @justsml in #74
- Search caches: bypass only in strict-privacy, tag spans with privacy mode by @justsml in #80
- CI: routing eval, Node 20/22 package floor, worker-image step test (rest of archived #196) by @justsml in #81
- Preserve reasoning across chat agent steps (fix beam-1.0 400 on step 2) by @justsml in #82
- Record zero-result diagnostics after the search session row commits by @justsml in #83
- Stop the DB watchdog from killing a saturated app; bound search SQL by @justsml in #84
- Add bounded document summaries and workspace summary option by @justsml in #89
- Chat agent turn deadline, turn-scoped cancellation and disconnect-safe request forwarding by @justsml in #91
- Refund chat turns whose answer never reached the client by @justsml in #90
- Bound document summaries with measured sampling and output limits by @justsml in #93
- Search: LanceDB-native retrieval by default, fair admission, batch embeddings by @justsml in #92
- Raise chat agent defaults; plan-based rate limits, higher Free limits and upgrade links on 402/429 by @justsml in #94
- Generate and upload the Stripe Agentic Commerce catalog feed by @justsml in #95
- Search: drop the per-search chunk liveness query from native hydration by @justsml in #96
- Add DocSlurp Claude plugin and harden CI failure handling by @justsml in #97
- Refresh dependencies and retire Docling for Pillow security fixes by @justsml in #98
- One workspace field (ID, slug or name) on every request by @justsml in #99
- Credit packs and per-unit pricing; remove plans and subscriptions by @justsml in #100
- Give the DocSlurp plugin a listing icon and explicit policy URLs by @justsml in #103
- Tuned DeepSeek V4.1 Flash 1017 as the default vision model by @justsml in #102
- List third-party data providers in the DocSlurp plugin docs by @justsml in #104
- Give MCP tools, prompts and the server directory-ready titles and metadata by @justsml in #105
- Harden API authorization, request bounds and stored-secret rotation by @justsml in #106
- Close round-2 security findings and repair outbound HTTPS fetches by @justsml in #107
- Send vetted outbound requests through undici's own fetch by @justsml in #108
- Google sign-in: account picker; strip sign-in tokens on repeat sign-in by @justsml in #111
- Built-in Google OIDC hosts; default-workspace viewers can upload by @justsml in #110
- Fix chat retrieval, OAuth linking and ingestion security by @justsml in #109
- Tighten API-key reads, chat tool workspace scoping and model policy by @justsml in #112
- Fix chat transcription 502s and stale/unredacted summary evidence by @justsml in #113
- Add administrator credits and accelerate exact board byte totals by @justsml in #116
- Render PDF pages on demand and repair ingestion and credit recovery by @justsml in #118
- Sharpen selected PDF pages and add bounded PNG export renders by @justsml in #119
- Finalize uploads automatically with multipart reconciliation and Tigris webhooks by @justsml in #117
- Notify operators about vendor billing failures, not customer credits by @justsml in #120
- Keep page previews fast and render sharper exports on demand by @justsml in #121
- Let chat propose workspace settings and enable video by default by @justsml in #123
- Add web tools: provider-neutral web search, fetch, map, crawl and website ingest by @justsml in #122
- Ground agent capabilities and block failed tool turns by @justsml in #125
- Preserve blocked chat errors and ground capability guidance by @justsml in #126
- Stream web tool progress in chat status by @justsml in #124
- Make the API database pool survive a machine suspend/resume by @justsml in #127
- Fix activity slot starvation from the 5 s ramp throttle; allow a smaller floor worker by @justsml in #131
- Overlap lexical, page and Lance warm work with the query embedding by @justsml in #129
- Auto-recover non-fatal queue stops: supervised Temporal worker, retrying listeners, isolated sweeps by @justsml in #132
- Add embed throughput benchmark and opt-in length-sorted local embedding batches by @justsml in #133
- Re-measure PIPELINE_PROFILES worker seconds with provenance (pricing review) by @justsml in #130
- Ground chat upload guidance in workspace inventory; require per-item overview citations by @justsml in #134
- perf(db): cut worker Postgres round trips per page ~55-65% by @justsml in #135
- Log DB pool recycles only after suspend-length gaps by @justsml in #136
- Teach the DocSlurp plugin skill automatic uploads and web tools by @justsml in #137
- Gate web tools and chat behind their own API key scopes by @justsml in #140
- Fix heading paths and chunk context for converted HTML pages by @justsml in #138
- Direct chat answers: tool_choice none, agent.tools, parentRetrieval; fix final-step tool markup leak by @justsml in #139
- Stop /api/sso-lookup from revealing which mailboxes have SSO links by @justsml in #141
- Sign worker images and hash-pin worker Python dependencies by @justsml in #142
- Add least-privilege Postgres roles for the API, worker and backups by @justsml in #143
- Let ingest server keys reprocess, retag and delete their workspace content by @justsml in #144
- Give every workspace its own LanceDB database by @justsml in #145
- Require production mode for Telegram notifications by @justsml in #146
- Finish per-workspace LanceDB cleanup and maintenance by @justsml in #148
- Add suggestions mode and per-turn web toggle to the chat agent by @justsml in #147
- Make the chat agent truthful about settings; treat Stop as a cancel by @justsml in #149
- Read images through vision when OCR is unavailable by @justsml in #150
- Show real failure causes in the ingest report; let chat propose vision page selection by @justsml in #151
- Never suggest failed or replaced files as chat starters by @justsml in #152
- Stop run progress from failing on unlisted run event types by @justsml in #155
- Ignore injected document instructions quietly in chat by @justsml in #156
- Accept small inline files in the MCP upload tool by @justsml in #157
- Keep paid full indexing out of the chat agent's reach by @justsml in #159
- Make every feature available to every organization; price audio and video before upload by @justsml in #158
- Derive a friendly document title from content and expose it on the API by @justsml in #160
- Send each chat evidence passage to the model once by @justsml in #161
- Run workspace auto-names through the untrusted-title checks by @justsml in #162
- Cut chat cost: dynamic step budget, prompt caching, fewer tools, $0.015 answers by @justsml in #163
- Make the chat agent search before answering product questions by @justsml in #164
- Require agents to keep branches synced with their remotes by @justsml in #165
- Refresh official model prices and infrastructure observations by @justsml in #167
- Export documents and workspaces as organized zips by @justsml in #166
- Keep chat answering when a web search or fetch fails by @justsml in #168
- Bill GPT-6 Luna as standard and let chat fix rejected tool arguments by @justsml in #169
- List MCP tools in product order with consistent titles by @justsml in #171
- Price standard chat answers at 1 credit by @justsml in #170
- Route GLM/DeepSeek to their cheapest hosts; org switch for China-based providers by @justsml in #172
- Report native PDF engine use in stage stats, logs and traces; fail closed when required by @justsml in #173
- Keep the details support needs on spans by @justsml in #174
- Index search sessions by workspace; stop the integrity sweep's HEAD bursts by @justsml in #175
- Name the real caller of API queries, past the resilient pool's proxy by @justsml in #176
- Choose embedding pooling per route with ?pooling=mean|cls by @justsml in #177
- Publish config-change reindex runs so they start by @justsml in #178
- Prune index configs from the stored pipeline, never a stale snapshot by @justsml in #179
- Install poppler-data CMaps in worker images; fail prerequisites without them by @justsml in #180
- Clean up the public API: real /v1/chat/stream, delete uncalled routes, catalogue domain error codes by @justsml in #192
- Add worker/poppler data by @justsml in #196
- Escalate extraction on aligned page text-yield deficits by @justsml in #181
- Evaluate mixed PDF extraction yield with protected corpus splits by @justsml in #194
- Enforce company SSO on the server and revoke bypass sessions by @justsml in #195
- Keep active browser sessions signed in for 30 idle days by @justsml in #153
- Add extract mode: uploads that auto-delete after a retention window by @justsml in #185
- Apply existing bot CLA exemptions before secret access by @justsml in #199
- Refresh model and infra pricing data; mark prohibited models by @justsml in #184
- Bump github/codeql-action/upload-sarif from 4.38.3 to 4.39.0 in the github-actions group across 1 directory by @dependabot[bot] in #101
- Bound worker CI installs with an official Ubuntu mirror fallback by @justsml in #200
- Add docslurp login (browser + device OAuth), token forwarding, and CLI tokens on the REST API by @justsml in #188
- Guard the public repo against private pre-split history by @justsml in #186
- Make chat settings tools app-only; harden public chat limits by @justsml in #182
- Add passkeys and two-factor auth (Better Auth plugins) with an org MFA policy by @justsml in #193
- Correct canonical chat-stream URLs in session documentation by @justsml in #201
- Add in-app feedback reports with live high-severity alerts by @justsml in #189
- eval-sync: dual-target Langfuse sync for datasets, score configs and experiment runs by @justsml in #187
- Name untitled workspaces from uploads, summaries and the first question by @justsml in #154
- Add flag-gated MCP Apps upload widget and STORAGE_PUBLIC_ENDPOINT signing by @justsml in #114
- Separate API key controls by type (browser vs server) by @justsml in #191
- Chat instructions: fixed core rules, default template, scope decline, app-only Improve by @justsml in #190
- Add manual signing for the reviewed document VM image by @justsml in #203
- Run Doppler PDFs in isolated VMs with bounded page chunking by @justsml in #204
- Fix multipart uploads that repeat a terminal storage marker by @justsml in #205
- Prevent organization slug collisions during first sign-in by @justsml in #206
- Patch vulnerable transitive deps, refresh Node 24.22.1 images and minor/patch deps by @justsml in #208
- Add platform-admin user directory API with signup attribution and client surfaces by @justsml in #207
- Drop pgvector: build every database on stock Postgres by @justsml in #216
- Run the gated Postgres/Redis tests in CI and fix the tests that rotted by @justsml in #217
- Preserve safe error details and honor retry guidance by @justsml in #223
- Drop the legacy run-budget trigger left on restored databases by @justsml in #222
- Require fresh proof for sensitive account changes by @justsml in #225
- Verify test sessions before checking live OAuth key restrictions by @justsml in #226
- Add native magic-link sign-in with explicit confirmation by @justsml in #229
- Shard long documents across parallel VMs and worker slots, calibrated from observed timings by @justsml in #227
- Harden Better Auth recovery, breached passwords and auth rate limits by @justsml in #230
- Speed up routing probe, page rendering and vision reconciliation (output-neutral) by @justsml in #234
- Refresh minor/patch deps, Node 24.22.2 images, pnpm 10.34.9 by @justsml in #238
- Split PDFs over 10,000 pages across Doppler VMs; log every shard decision by @justsml in #231
- Remove the per-document vision page cap by @justsml in #233
- Fast / Auto / Good quality modes (default Auto) replace parser playbooks by @justsml in #236
- Fix legacy vision routing for scans, tables and math behind VISION_ROUTING_V2 by @justsml in #235
- Normalize vision whitespace and configure China-based provider defaults by @justsml in #237
- Preserve HTML table chunks and scan nested repository integrity by @justsml in #240
- Fix nested chat failures and bound first-token waits by @justsml in #242
- Add security-council follow-up tickets from the live Ask AI assessment by @justsml in #209
- Harden upload admission and browser instruction telemetry (partial follow-ups) by @justsml in #244
- Persist scoped workspace recommendation snapshots and acknowledgements by @justsml in #246
- Classify generic quantity overlaps using parsed comment context by @justsml in #245
- Support exact-scan operator adjudication of integrity candidates by @justsml in #248
- Preserve visual relationships and uncertainty in vision instructions by @justsml in #247
- Preserve embedded PDF images as lossless WebP in the Doppler VM by @justsml in #243
- Compact faithful oversized table chunks by @justsml in #241
- Keep hashed email codes valid for 12 hours; replace on resend by @justsml in #239
- Preserve table spans and gate competitive cell coverage by @justsml in #250
- Make MMR diversify results; remove benchmark-specific extraction rewrites by @justsml in #249
- Document the mmr search option; date the old MMR eval by @justsml in #251
- Correct raster quarter turns with an isolated CPU model by @justsml in #252
- Enable CPU image orientation by default by @justsml in #257
- Add geo contracts and preserve structured extraction annotations by @justsml in #260
- Upgrade LanceDB JavaScript and Python to 0.41 by @justsml in #263
- Integrate geo final-source authority (work in progress) by @justsml in #262
- Add opt-in Speed boost for chat and ingestion with 4x credits by @justsml in #264
- Checkpoint: fail-closed native capabilities and geo predicate builders by @justsml in #265
- Fix migration owner preflight for installed PostGIS by @justsml in #266
- Fall back to the base model when a Nitro route returns 404 by @justsml in #267
- Retry graph fallbacks and extract documents in parallel by @justsml in #269
- Use NFC/NFD instead of NFKC/NFKD for text normalization by @justsml in #271
- Keep runs waiting for their next activity in scaler demand by @justsml in #272
- Give run-level stages the run's provider scope; start text calls at the lowest reasoning by @justsml in #270
- Log boolean env flags at worker and API startup by @justsml in #274
- Offer click-through chat recovery for empty retrieval indexes by @justsml in #275
- Correct database health alert provider and failure descriptions by @justsml in #276
- Make forced Temporal worker shutdown process-fatal by @justsml in #277
- Upgrade compatible dependencies, including WebAuthn 14 and TypeScript 7 by @justsml in #280
- Use the maskable app icon as the favicon and primary icon by @justsml in #281
- Correct and fill gaps in public docs against current code by @justsml in #282
- Add session realtime projections and native Mastra operation recovery by @justsml in #278
- Preserve schema owner through coordinated superuser migrations by @justsml in #286
- Reject unsupported Algolia-compat filters instead of ignoring them by @justsml in #283
- Require chat scope for aggregate answers; accept workspaces:manage keys on search settings by @justsml in #284
- Keep strict-privacy searches out of the search response cache by @justsml in #285
- Release v0.4.0 by @justsml in #287
New Contributors
- @dependabot[bot] made their first contribution in #101
Full Changelog: v0.3.0...v0.4.0