Changelog

All notable changes to GNO will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

0.41.1 - 2026-04-07

Fixed

• Fixed sqlite-vec writes to replace existing chunk rows cleanly during re-embed flows, eliminating the duplicate-key warning users could hit after same-model clears or force re-embeds.
• Fixed embedding recovery so GNO now downshifts failed batches, resets disposed embedding ports, and surfaces concrete sample errors plus retry guidance instead of only reporting opaque failed-chunk counts.

0.41.0 - 2026-04-07

Added

• Added embedding compatibility profiles plus safer indexing fallback behavior, including model-aware query/document formatting, batch-to-single-item embed recovery, and batched vector-query embedding in hybrid retrieval.

Changed

• Improved the default Qwen embedding path for both indexing and retrieval while keeping the full retrieval stack intact, and documented the required gno embed --force or clear-and-reembed migration path for formatting/profile changes.

0.40.2 - 2026-04-06

Fixed

• Fixed Web UI document layout regressions in the left outline rail, including width blowouts, sticky behavior, and long-title handling in dense note views.
• Fixed local markdown image rendering in Doc View and editor preview by resolving note-relative asset paths against the current document instead of the web root.

0.40.1 - 2026-04-06

Changed

• Updated the release workflow to use actions/download-artifact@v8 in publish.yml.

0.40.0 - 2026-04-06

Added

• Added collection-scoped embedding ergonomics across CLI, Web UI, API, and MCP, including gno embed [collection] / gno embed --collection <name> plus collection embedding cleanup actions for stale or all vectors.
• Added gno_clear_collection_embeddings to MCP so agents can explicitly remove stale or all embeddings for one collection when write tools are enabled.

Changed

• Updated CLI, MCP, Web UI, and skill docs so collection-scoped embed/re-embed/cleanup flows are documented consistently after the default embed-model switch.

0.39.1 - 2026-04-06

Changed

• Clarified the upgrade path after the default embed-model switch to Qwen3-Embedding-0.6B-GGUF, including the explicit gno models pull --embed + gno embed re-embed step for existing collections.

0.39.0 - 2026-04-06

Added

• Added a new public multilingual markdown benchmark lane for general collections, with vendored FastAPI docs fixtures, same-language and cross-language relevance cases, and committed benchmark artifacts for bge-m3 and Qwen3-Embedding-0.6B-GGUF.
• Added collection-level embedding cleanup across CLI, Web UI, and API, including stale-vs-all modes and protection for vectors still shared by active documents in other collections.

Changed

• Switched built-in preset embedding models from bge-m3 to Qwen3-Embedding-0.6B-GGUF after it materially outperformed bge-m3 on both code and multilingual prose benchmark lanes.
• Status/backlog reporting is now model-aware for the active embedding model across CLI, Web UI, SDK, and MCP, so preset/default embed-model changes immediately surface the need to re-embed.
• Preset switching in CLI/Web/API now explicitly tells users when the embedding model changed and embeddings need to be regenerated.

0.38.0 - 2026-04-06

Added

• Added a collection-level model editor in the web UI, including per-role override editing, inherited-vs-overridden effective model display, and code-collection recommendations for the embedding model.
• Added richer collection API payloads plus PATCH /api/collections/:name so clients can inspect and update collection-scoped model overrides without hand-editing YAML.
• Added gno collection add --embed-model so collection-specific embedding model overrides can be set directly from the CLI.

Changed

• Updated CLI, API, configuration, Web UI, troubleshooting, and agent-skill docs to cover collection-scoped model overrides and the follow-up embedding workflow after changing a collection’s embedding model.

0.37.0 - 2026-04-06

Added

• Added a retrieval-quality upgrade pass across BM25 lexical handling, code-aware chunking, terminal result hyperlinks, and per-collection model overrides.
• Added a code-embedding benchmark workflow with canonical, real-GNO, and pinned OSS benchmark slices plus a bounded autonomous search harness for comparing alternate embedding models.
• Added benchmark/result pages and documentation that now recommend Qwen3-Embedding-0.6B-GGUF as the current code-specialist embedding model for per-collection code overrides.
• Added ADRs for BM25 query semantics, code-aware chunking, and collection-model resolution, plus a follow-up Flow epic for more granular path/file-type model resolution.

Fixed

• Fixed BM25 lexical search so quoted phrases, negation, hyphenated technical terms, underscores, and filepath/title/body weighting behave intentionally and are protected by regression tests.
• Fixed CLI retrieval output to emit TTY-only OSC 8 hyperlinks without leaking escape sequences into structured or non-TTY output.
• Fixed code-chunking visibility so gno doctor now reports the automatic code-aware chunking mode and supported extensions.
• Fixed the embedding autoresearch harness so candidate runtime wiring, dual-fixture scoring, and leaderboard/result loading work correctly in the tracked branch/CI path.

0.36.0 - 2026-04-06

Added

• Added a much stronger note command palette experience in the web workspace, including place-aware creation, section-first query results, keyboard navigation that follows visible ordering, and better help/discoverability around note-native actions.

Changed

• Rebalanced the desktop document view rails so metadata and outline stay together on the left, while properties/path and relationship panels live on the right with tighter sticky behavior for long-note reading.

Fixed

• Fixed command palette focus/selection styling so the input no longer inherits the global outline ring and selected items are visually obvious while using arrow keys or narrowing results.
• Fixed command palette result ordering so cmdk no longer fights the app’s grouped ordering, preventing invisible selection jumps to non-visible items.

0.35.0 - 2026-04-05

Added

• Added workspace-native note authoring flows across the web workspace, including place-aware note creation from Browse, note presets, editor preset insertion, section outline/deep links, and a stronger command palette for note/navigation actions.
• Added reference-aware file operations for editable notes, including move, duplicate, folder creation, refactor warning previews, and shared note/file operation contracts that now span the Web UI, SDK, and MCP surfaces.
• Added new programmatic surfaces for note workflows: REST endpoints for note presets, sections, folder creation, file refactor planning, move, and duplicate; new MCP write tools for folder/note refactors; and SDK parity for note creation, folder creation, sections, rename, move, and duplicate.
• Added the Scholarly Dusk design-system ADR and linked the new note-workspace epics/tasks to it so future UI work stays visually coherent.

Fixed

• Fixed Windows/macOS path normalization issues so note/file workflows and desktop runtime-layout helpers keep POSIX-stable relpaths/URIs across platforms, unblocking green Windows CI again.

0.34.1 - 2026-04-04

Fixed

• Fixed the desktop release workflows to stop uploading redundant proof-only artifacts, preventing transient GitHub artifact ENOTFOUND failures from sinking otherwise successful macOS packaging runs.

0.34.0 - 2026-04-04

Added

• Added a cross-collection Browse 2.0 workspace with a real tree sidebar, folder detail panes, and tab-scoped browse state restoration.
• Added richer document rendering in the web workspace, including clickable frontmatter links, clickable resolved wiki links in note content, and denser document metadata/frontmatter presentation.
• Added a broad website/content refresh across the homepage, feature pages, docs metadata, FAQ, and comparison pages to better position GNO as a local knowledge workspace for humans and agents.

Fixed

• Improved graph loading by reducing redundant server-side graph work and tuning the client-side graph defaults for faster initial render.
• Improved browser smoke coverage for Browse flows and stabilized the direct-child browse API assertion in tests.

0.33.4 - 2026-04-03

Fixed

• Improved web UI regression coverage with Bun-first DOM interaction tests and a repo-local browser smoke path for search, capture, and document flows.

0.33.3 - 2026-04-03

Fixed

• Fixed the macOS desktop packaging workflow artifact handling so the release-manifest JSON is uploaded from the generated release directory instead of the wrong artifact root.

0.33.2 - 2026-04-03

Fixed

• Fixed the macOS desktop packaging workflow to execute the release script directly in CI and validate the artifact directory in the same step, reducing ambiguity around artifact discovery failures.

0.33.1 - 2026-04-02

Fixed

• Fixed the macOS desktop release workflow so notarized desktop beta artifacts are uploaded from the correct runner paths and can be attached to GitHub releases.

0.33.0 - 2026-04-02

Added

• Added a packaging matrix for CLI vs desktop support on macOS and Linux, including support tiers, artifact guidance, and runtime bundling assumptions.
• Added a repo-local macOS desktop release command for the Electrobun shell that signs, notarizes, staples, and verifies versioned desktop beta zip/DMG artifacts.
• Added a macOS desktop packaging job to the release workflow so tagged releases can build notarized desktop beta artifacts in CI once the release environment is configured.

Changed

• Installation and desktop rollout docs now describe Linux desktop as experimental and macOS desktop as the primary beta target.

0.32.0 - 2026-04-02

Added

• Added GET /api/jobs/active so clients can discover the currently running background job without scraping a 409 error message.

Changed

• Job-conflict 409 API responses now include structured error.details.activeJobId metadata.

0.31.2 - 2026-04-02

Fixed

• Browse now uses the available table space more effectively for long note titles, paths, and collection names. Document titles/paths wrap cleanly, and collection chips no longer truncate similarly named collections into unreadable slivers.

0.31.1 - 2026-04-01

Fixed

• Relaxed Windows-only CI assumptions in the desktop runtime-layout tests so path separator differences no longer fail the test suite on win32.
• Increased the Windows timeout budget for the concurrent CLI access regression test, avoiding false negatives from slower process startup on GitHub Actions runners.

[0.31.0] - 2026-04-01

Added

• Added first packaged windows-x64 desktop beta build plumbing, including packaged-runtime validation, Windows packaging workflow coverage, and release-asset support for the Windows desktop zip.
• Added explicit Windows support docs covering current target scope, packaged-runtime validation, and manual validation guidance for the desktop beta.

Changed

• gno index <collection> now scopes the embedding phase to that collection instead of consuming unrelated global backlog from other collections.
• Release automation now publishes the packaged Windows desktop beta zip alongside the normal npm/GitHub release flow.

Fixed

• CLI index summaries no longer divide embed duration by 1000 twice, so long embedding runs report realistic times instead of bogus sub-second output.

0.30.0 - 2026-03-27

Added

• Added gno daemon, a headless continuous-indexing mode that reuses GNO’s existing watcher/sync/embed pipeline without starting the Web UI server.

Changed

• Elevated the in-app workspace presentation across the web/desktop shell path with stronger visual hierarchy, cleaner tabs/footer treatment, and a more intentional dashboard/search/ask aesthetic.
• Clarified product docs around when to use the desktop app, gno serve, and the new headless daemon mode.

Fixed

• Read-only CLI commands no longer take unnecessary write locks on startup, avoiding transient database is locked failures when they overlap with gno update.
• gno daemon --no-sync-on-start now behaves correctly and skips the initial sync pass.

0.29.2 - 2026-03-27

Changed

• Refined the gno serve workspace presentation with a more distinctive visual language across the dashboard, tabs, search, ask, footer, and global styling, while keeping the UI fully offline-safe.

0.29.1 - 2026-03-27

Fixed

• Desktop/Web document trashing no longer depends on a separately installed global trash CLI. GNO now uses built-in platform-aware trash behavior instead.
• Trashing a document now reports failure if the document cannot be marked inactive in the index, avoiding false-success UI states that could leave stale search results behind.

0.29.0 - 2026-03-27

Added

• First public GNO Desktop Beta shell naming and rollout surfaces for the mac-first desktop app path.
• Full desktop-beta onboarding flow in the Web UI and desktop shell, including folder setup, plain-language preset selection, health checks, bootstrap/runtime visibility, connector center, import preview, app tabs, file lifecycle actions, and recovery history.
• Explicit in-wizard sync progress plus clearer blocked-state handling for first indexing and embedding completion.

Changed

• slim-tuned is now the built-in default preset, using the fine-tuned expansion model while keeping the same embed, rerank, and answer models as slim.
• Fresh onboarding now skips the model-prep step when the active preset is already ready.
• Desktop shell now exposes standard edit menu actions so paste/select-all work like a normal app.

Fixed

• Collection add no longer throws a false “Collection not found after add” error after successful mixed-case name normalization.
• Fresh sandboxes and alternate runtime launches no longer mis-detect model readiness when GNO_CACHE_DIR already points at a models directory.
• No-op Update All runs now report a stable up-to-date result instead of flashing away.
• Onboarding sync/add flows now trigger embedding immediately when models are ready, preventing the final step from looking stuck after indexing starts.

0.28.2 - 2026-03-23

Fixed

• Updated the validated dev-tooling dependency batch: @biomejs/biome, @types/react, lefthook, oxlint-tsgolint, playwright, and ultracite.

0.28.1 - 2026-03-23

Fixed

• Updated @modelcontextprotocol/sdk to 1.27.1 in a narrow follow-up security/maintenance bump after the stale grouped dependency PR failed local validation.

0.28.0 - 2026-03-23

Fixed

• gno embed now requests full embedding threads from node-llama-cpp and uses an adaptive embedding-context pool on CPU-only machines, with graceful fallback when extra contexts cannot be created. Thanks @riyadist for the report and repro details.

0.27.3 - 2026-03-23

Fixed

• Excluded the archived desktop/electrobun-spike workspace from root type-aware lint/typecheck so CI and patch publishes no longer fail on unresolved spike-only Electrobun imports.

0.27.2 - 2026-03-23

Fixed

• Fell back gracefully for unsupported markdown fence languages in the Web UI, including Obsidian-style ````tasks` blocks that previously crashed document rendering with a frontend ShikiError. Thanks @almino for the report.

0.27.1 - 2026-03-22

Fixed

• Normalized created gno:// document URIs to POSIX-style forward slashes on Windows so editable-copy responses and related create flows no longer return backslash-separated URIs.

0.27.0 - 2026-03-22

Added

• Safe document capability metadata across the API, CLI gno get, SDK, and MCP so callers can distinguish editable markdown/plaintext documents from read-only converted source files.
• Read-only handling plus editable-copy creation for converted PDF/DOCX-style documents in the Web UI.
• Deep links for document view/edit routes with source-view line targeting and exact-hit navigation from search results.
• Watch-driven document event streaming for live refresh and external-change awareness in Search, Browse, Doc View, and the editor.
• Wiki-link autocomplete and linked-note creation in both the full editor and Quick Capture.
• A fast Cmd/Ctrl+K quick switcher with recent-document tracking and note-creation handoff.
• Local editor snapshot history with restore support after save conflicts or mistaken edits.
• Updated skill guidance for editable vs read-only document handling and richer gno get --json capability metadata.

Changed

• Web/docs/website positioning now describe GNO as a safer markdown-first local workspace for agent-centric teams rather than only a retrieval layer alongside Obsidian.
• gno_get / GET /api/doc / related docs now return richer source metadata and capability metadata.

Fixed

• Preserved deep-link line targets when navigating from document view into edit mode.
• Avoided editable-copy navigation races by waiting for the copied markdown document to become queryable before opening it.
• Removed the ugly quick-switcher focus ring/outline clash from the command input.

0.26.0 - 2026-03-22

Added

• OpenCode and OpenClaw as skill install targets (--target opencode, --target openclaw).
• --target all now installs to all 4 targets (claude, codex, opencode, openclaw).
• MCP parameter descriptions on all 19 tools, including clearer usage guidance across roughly 60 parameters.
• MCP tool descriptions with improved usage guidance, async/job polling hints, and ref-format examples.
• Skill documentation for --since, --until, --exclude, --intent, and --query-mode.
• Skill sections for Document Retrieval, Links & Similarity, and search-then-get JSON pipelines.

Fixed

• --target all for skill install/uninstall was previously hardcoded to claude + codex only.

0.25.1 - 2026-03-22

Changed

• Expanded the bundled GNO agent skill with missing CLI retrieval flags, structured query mode guidance, get/multi-get retrieval examples, search-then-get JSON pipelines, and document links/similarity commands for better agent usability.

[0.25.0] - 2026-03-21

Added

• Preset-aware retrieval depth policy across Web, CLI, and MCP: Balanced now enables expansion on slim / slim-tuned, while Thorough widens the rerank candidate pool for best recall.
• Explicit model-role split for presets and runtime: expand now controls query expansion separately from gen answer generation, with updated config/docs, model management UX, and CLI output.
• Collection management discoverability and actions across the web UI, including direct /collections navigation, inline reindex affordances, and clickable collection surfaces that lead into filtered Browse views.

Changed

• Ask/Search preset UI now presents the active preset as a bundle and shows separate expansion vs answer roles instead of implying the retrieval-tuned model is the answer model.
• The live slim-tuned preset guidance now pairs the promoted retrieval expansion model with a separate larger answer model.

[0.24.0] - 2026-03-10

Added

• First-class structured multi-line query documents using term:, intent:, and hyde: across CLI query/ask, REST /api/query and /api/ask, MCP gno_query, SDK query/ask, and Web Search/Ask text boxes.
• Dedicated structured syntax reference doc plus updated CLI/API/MCP/SDK/Web docs.

[0.23.0] - 2026-03-10

Added

• First stable SDK / library mode at the package root via createGnoClient(...), with inline-config and file-backed startup, direct retrieval/document methods, and programmatic update / embed / index flows.

Changed

• Package root now resolves to the SDK surface; the CLI remains available through the gno binary and ./cli export.
• Added full SDK docs, README coverage, architecture notes, website nav, homepage copy, and a dedicated SDK feature page.

[0.22.6] - 2026-03-10

Fixed

• Stabilized cross-platform research tests by replacing one remaining generated-artifact assertion with direct filtering logic checks.

[0.22.5] - 2026-03-10

Changed

• Updated the main README, CLI examples, docs page, homepage bento, and fine-tuned model feature page to point at the current promoted slim-tuned retrieval model and Hugging Face install flow.

[0.22.4] - 2026-03-10

Changed

• Updated the public fine-tuned model docs and feature pages to point at the published Hugging Face model, the current promoted slim retrieval preset, and the canonical HF-backed install flow instead of local-only paths.

[0.22.3] - 2026-03-10

Fixed

• Relaxed the autonomous confirmation test so release CI no longer depends on a locally generated repeat-benchmark artifact being committed.

[0.22.2] - 2026-03-10

Fixed

• Made retrieval research tests portable in CI by removing machine-specific absolute paths and using repo-root-relative fixture resolution.

[0.22.1] - 2026-03-10

Fixed

• Included fine-tuning helper libraries and benchmark fixture JSON used by the retrieval research scripts/tests so release CI can resolve the research sandbox correctly on clean machines.

[0.22.0] - 2026-03-10

Added

• Retrieval fine-tuning sandbox now supports real local MLX LoRA training, portable GGUF export, automatic checkpoint selection, promotion bundles, and repeatable benchmark comparisons.
• Autonomous retrieval search harness now supports bounded candidate search, early-stop guards, repeated incumbent confirmation, promotion target checks, and recorded experiment history.
• Published promoted slim retrieval model gno-expansion-slim-retrieval-v1 on Hugging Face, plus canonical release bundle, install snippet, and user-facing fine-tuned model documentation.

[0.21.1] - 2026-03-09

Added

• Reproducible retrieval candidate benchmark harness for local generation bases, including Qwen3.5 comparisons, raw artifacts, and recommendation memo for expansion fine-tuning.
• Manual benchmark scripts/docs for full-path model evaluation across expansion reliability, hybrid retrieval quality, latency, memory, and ask-style smoke tests.

[0.21.0] - 2026-03-08

Added

• CLI gno ask --query-mode parity with existing Ask API and Web support, including structured mode validation and JSON metadata output.

[0.20.0] - 2026-03-08

Changed

• Upgraded node-llama-cpp to 3.17.1 and switched runtime initialization to build: "autoAttempt" for improved backend selection and fallback behavior.

[0.19.0] - 2026-03-08

Added

• Explicit exclusion filters across CLI, API, Web, and MCP retrieval surfaces for hard-pruning docs by title/path/body terms.
• Ask-side structured query mode parity across API and Web, including validated queryModes pass-through and Ask UI chips.

Changed

• Ask responses now surface structured query mode summary metadata for retrieval debugging and UX confirmation.

[0.18.0] - 2026-03-08

Added

• Explicit intent steering for ambiguous retrieval across CLI, API, Web, and MCP query flows.
• candidateLimit controls for hybrid retrieval and ask flows to tune rerank cost vs. recall.

Changed

• Query expansion now uses a bounded configurable generation context (models.expandContextSize, default 2048).
• Reranking now deduplicates identical chunk texts before scoring and fans scores back out deterministically.
• Search and Ask web advanced retrieval controls now expose intent and rerank candidate limit.

[0.17.0] - 2026-02-23

Added

• Structured query mode inputs across CLI/API/MCP (term, intent, hyde) with validation and explain metadata.
• Temporal retrieval upgrades: query recency intent detection, explicit/relative date range parsing, and recency sorting with frontmatter-date fallback.
• Frontmatter date-field extraction and date-aware browse sorting (sortField/sortOrder) in web API and UI.
• Logseq link compatibility for alias links ([text]([[Target]])) and block embeds ({{embed ((block-id))}}).

Changed

• Ingestion now stores richer metadata/date materialization in documents for retrieval-time filtering and sorting.
• Web routing now remounts pages on URL changes to keep Browse/Doc views reactive without custom navigation events.
• Documentation/specs refreshed for query modes, temporal filters, API sorting, and retrieval behavior.

Fixed

• Max-bytes enforcement now re-checks file size before read/convert to prevent stale-walker oversize ingestion.
• Link extraction no longer double-counts alias inner wiki syntax in Logseq-style links.

Migrations

• Added migration 006-document-metadata (documents.metadata_json).
• Added migration 007-document-date-fields (documents.document_date, documents.date_fields_json) with backfill support.

[0.16.0] - 2026-02-03

Changed

• Improved batch embedding performance - embedBatch() now processes embeddings concurrently (up to 16 at a time) instead of sequentially, with proper error handling and dispose safety (#64)

[0.15.1] - 2026-02-01

Changed

• Upgrade officeparser 5.2.2 → 6.0.4 (v6 API: parseOffice().toText())
• Upgrade 23 dependencies including oxlint 1.42.0, oxfmt 0.27.0, node-llama-cpp 3.15.1

Fixed

• Await pager stdin write (caught by new oxlint rule)

[0.15.0] - 2026-02-01

Added

• HTTP backends for remote model servers - Offload embedding, reranking, and generation to remote llama-server instances via OpenAI-compatible APIs (#62) - thanks @Whamp!
  • Configure with URI format: http://host:port/path#modelname
  • Supports embedding (/v1/embeddings), reranking (/v1/completions), and generation (/v1/chat/completions)
  • Enables running GNO on lightweight machines while GPU inference runs on separate servers

[0.14.3] - 2026-01-14

Changed

• CLI embed verbose errors now include batch titles and root cause

[0.14.2] - 2026-01-14

Added

• Frontmatter metadata grid in doc view and edit preview

Fixed

• Strip frontmatter from markdown preview rendering
• Prevent frontmatter URL overflow in UI
• Stop dev CSS flicker by relying on built stylesheet

[0.14.1] - 2026-01-14

Added

• Verbose embedding errors - --verbose flag now logs embedding failures to stderr
  • gno index --verbose and gno embed --verbose show batch failures, count mismatches, and store errors
  • Helps debug when gno index reports errors without details

[0.14.0] - 2026-01-10

Changed

• Walker include fallback - Empty include now defaults to supported document types (.md, .txt, .pdf, .docx, .pptx, .xlsx) instead of all files
  • Prevents “No converter for application/octet-stream” errors when indexing repos with source code
  • Extensionless files (Makefile, LICENSE) and dotfiles (.env, .gitignore) are always excluded
  • Explicit include still overrides the default

Fixed

• Documentation now accurately describes exclude patterns as component-based matching (not globs)

0.13.2 - 2026-01-07

Fixed

• Disable pager in tests and when NO_PAGER/GNO_NO_PAGER is set

Changed

• Clarify release workflow changelog handling

0.13.1 - 2026-01-07

Fixed

• Escape backslashes in markdown table cell output

0.13.0 - 2026-01-06

Added

• Knowledge Graph - Interactive visualization of document connections
  • gno graph - CLI command with --json, --dot, --mermaid output formats
  • --collection filter, --similar for similarity edges, --threshold control
  • REST API: GET /api/graph with full query parameters
  • MCP tool: gno_graph for AI agents to explore relationships
  • WebUI: /graph page with force-directed layout, zoom/pan, collection filter
  • Similarity edges shown as golden connections (semantic relatedness)
  • Click any node to navigate to that document
• Document Viewer - Enhanced document reading experience
  • Outgoing links panel showing wiki and markdown links
  • Backlinks panel showing documents linking to current doc
  • Related notes sidebar with AI-powered similarity suggestions
  • Tooltips for truncated titles in all panels

Fixed

• Wiki link resolution for path-based references
• Right panel title truncation with tooltip on hover

0.12.0 - 2026-01-05

Added

• Note Linking - Wiki-style links, backlinks, and semantic similarity
  • [[Target]] wiki link syntax with cross-collection support ([[collection:Target]])
  • gno links <docid> - List outgoing links from a document
  • gno backlinks <docid> - Find documents linking TO a target
  • gno similar <docid> - Semantic similarity search using embeddings
  • REST API: /api/doc/:id/links, /api/doc/:id/backlinks, /api/doc/:id/similar
  • MCP tools: gno_links, gno_backlinks, gno_similar
  • WebUI: OutgoingLinksPanel, BacklinksPanel, RelatedNotesSidebar
  • WikiLinkAutocomplete: [[ trigger, fuzzy search, keyboard nav, broken link detection
• Vector Index Maintenance - Fix empty results from gno similar
  • gno vec sync - Fast incremental sync when vec0 drifts
  • gno vec rebuild - Full rebuild of vec0 index
  • Auto-sync after embed batches when drift detected
  • Transaction-wrapped sync for atomicity

Fixed

• vec0 index sync - Vector inserts silently failing no longer leaves index out of sync
• inferDimensions - Filter by model and validate byte alignment

0.11.0 - 2026-01-05

Added

• MCP gno_embed tool - Generate embeddings for unembedded chunks via MCP
  • Runs as async background job, poll with gno_job_status
  • Offline-only: fails fast if embedding model not cached (no auto-download)
  • Returns { jobId, status, model } on start
• MCP gno_index tool - Full reindex (sync + embed) in single job
  • Syncs all collections then embeds new chunks
  • Optional collection param to limit scope
  • Optional gitPull to pull before sync
  • Returns { jobId, collections, phases, status }
• Typed job results - Job status now includes discriminated typedResult union
  • { kind: "sync", value: SyncResult }
  • { kind: "embed", value: { embedded, errors } }
  • { kind: "index", value: { sync, embed } }
• Server-side embed scheduler - Auto-embed after sync in web UI
  • 30s debounce to batch rapid syncs
  • Background embedding with progress in /api/embed/status
  • Manual trigger via POST /api/embed

Fixed

• Embed scheduler notification - Sync/add jobs now properly trigger auto-embed
• Job schemas - Updated mcp-job-status.schema.json and mcp-job-list.schema.json for new job types

0.10.4 - 2026-01-04

Changed

• Skill progressive discovery - Concise SKILL.md overview with links to reference docs (cli-reference.md, mcp-reference.md, examples.md) for reduced context bloat

0.10.3 - 2026-01-04

Fixed

• Markdown title extraction - Files without # heading now fall back to filename instead of showing URL-encoded path in UI

0.10.1 - 2026-01-04

Fixed

• Tailwind v4 semantic colors - Add @theme directive to generate bg-card, bg-muted, text-foreground etc. utilities (previously silently failing)
• Tag autocomplete - Substring matching (typing “fix” now finds “fixtures”), retry on fetch failure, re-filter when tags load
• POST /api/docs URI - Return gno:// URI instead of file:// for correct navigation
• CodeMirrorEditor - Migrate to React 19 ref-as-prop pattern

0.10.0 - 2026-01-04

Added

• Tag System - Organize and filter documents by tags
  • Frontmatter tag extraction during sync (tags: field in markdown)
  • Hierarchical tags with / separator (e.g., project/web, status/draft)
  • gno tags CLI commands: list, add, rm for tag management
  • --tags-any/--tags-all flags on search/vsearch/query/ask for OR/AND filtering
  • REST API: GET /api/tags endpoint, tag filtering params on search endpoints
  • MCP tools: gno.list_tags for tag discovery, gno.tag for add/remove
  • WebUI: TagInput with autocomplete, TagFacets sidebar for filtering, filter chips display

0.9.6 - 2026-01-03

Fixed

• SKILL.md frontmatter - allowed-tools now space-delimited per Agent Skills spec

0.9.5 - 2026-01-03

Added

• --skill flag for agent discovery - gno --skill outputs SKILL.md for AI agents to discover capabilities
  • Follows Agent Skills specification format
  • Clean output suitable for piping/parsing

Fixed

• SKILL.md refresh - Updated with ~15 missing CLI commands (serve, models, context, mcp, skill subcommands, doctor, etc.)

0.9.4 - 2026-01-03

Changed

• Default model preset switched to “slim” - Better eval scores (69% vs ~50%), faster inference, smaller download (~1GB)
  • Balanced preset still available via gno models use balanced
• Updated model names - slim: Qwen3-1.7B, balanced: Qwen2.5-3B-Instruct, quality: Qwen3-4B-Instruct

Internal

• Evalite v1 evaluation harness (local-only quality gates for releases)
• Fixed --full flag ranking order bug in search results

0.9.3 - 2026-01-03

Added

• Shell tab completion - bash/zsh/fish support with commands, flags, and dynamic collection names
  • gno completion output <shell> - Generate completion script
  • gno completion install - Auto-install to shell config
• Automatic pager - Long output piped through $PAGER (less -R / more)
  • --no-pager flag to disable
  • Respects TTY detection and structured output modes

0.9.2 - 2026-01-03

Added

• ThoroughnessSelector on Ask page - Fast/Balanced/Thorough modes with keyboard shortcut (T)
• Skills documentation - New docs/integrations/skills.md with Claude Code, Codex, OpenCode, Amp, VS Code Copilot, and Cursor support
• noExpand/noRerank API params - Added to /api/ask endpoint for search depth control

Changed

• Documentation restructure - Agent integration (Skills, MCP) now prioritized before Web UI in README and website
• Screenshot updates - All webui screenshots updated to .jpg format with new Brandbird styling

Fixed

• react-markdown node prop leak - Prevented node prop from leaking to DOM elements
• Table row striping - Increased visibility of alternating row colors in markdown tables
• Em dash removal - Replaced AI-style em dashes with standard punctuation across all docs

0.9.1 - 2026-01-02

Added

• Document Editor Sync Scrolling - Bidirectional scroll sync between CodeMirror editor and markdown preview
  • Toggle button (Link/Unlink icons) in toolbar, enabled by default
  • Percentage-based position mapping with event-based loop prevention
  • Epsilon checks prevent unnecessary updates and jitter

Fixed

• Raycast MCP docs - Fixed deeplinks with correct mcpServers wrapper format, added clipboard auto-fill JSON

0.9.0 - 2026-01-02

Added

• MCP Write Operations - AI assistants can now manage collections via MCP
  • gno_capture - Create new markdown documents in collections
  • gno_add_collection - Add folders to the index (async with job tracking)
  • gno_sync - Reindex one or all collections
  • gno_remove_collection - Remove collection from config
  • gno_job_status - Check async job progress
  • gno_list_jobs - List active and recent jobs
• Write Tool Gating - Disabled by default, enable with --enable-write or GNO_MCP_ENABLE_WRITE=1
• Security Protections
  • Dangerous root rejection (/, ~, /etc, ~/.ssh) with realpath canonicalization
  • Path traversal prevention (rejects ../ escapes)
  • Sensitive subpath blocking (.ssh, .gnupg, .git)
  • Cross-process locking via OS-backed flock/lockf
  • Atomic writes (temp + rename pattern)
• Core Modules - Shared utilities for MCP and WebUI
  • src/core/validation.ts - Path and collection validation
  • src/core/file-ops.ts - Atomic file operations
  • src/core/file-lock.ts - Advisory file locking
  • src/core/config-mutation.ts - Config change flow
  • src/core/job-manager.ts - Async job tracking
• MCP Documentation - Updated docs/MCP.md with all new tools, security model, Raycast write-enabled deeplinks
• Smoke Test Script - scripts/mcp-write-smoke-test.ts for MCP validation

0.8.6 - 2026-01-02

Added

• Single-key shortcuts (GitHub/Gmail pattern): N new note, / search, T cycle depth, ? help
• HelpButton with scholarly marginalia design (Old Gold accents)
• ShortcutHelpModal redesigned with two-column “Card Catalog” layout
• ThoroughnessSelector for search depth control (Fast/Balanced/Thorough)
• AIModelSelector with “vacuum tube display” aesthetic for Ask page
• Simplified Search page - Thoroughness now controls BM25 vs hybrid modes

Fixed

• Keyboard shortcuts no longer fire when Ctrl/Cmd/Alt held (fixes macOS browser conflicts)
• Shortcuts don’t fire inside dialogs or text inputs

0.8.5 - 2026-01-02

Fixed

• Pre-build Tailwind CSS for npm distribution - runtime plugin fails for global installs

0.8.4 - 2026-01-02

Fixed

• bunfig.toml missing from npm package - broke Tailwind CSS processing in gno serve

0.8.3 - 2026-01-02

Fixed

• Web UI shortcuts use Ctrl on all platforms (avoids Cmd+N/K browser conflicts on Mac)
• Missing minimatch dependency broke gno multi-get on npm install
• Removed unused rehype-highlight dependency

0.8.2 - 2026-01-02

Fixed

• gno serve broken when installed from npm - moved tailwindcss to dependencies

0.8.1 - 2026-01-02

Added

• AI Answers screenshot in docs (webui-ask-answer.png)

Fixed

• CI publish workflow: removed obsolete typecheck script, use lint:check
• Screenshot references updated to PNG format

0.8.0 - 2026-01-02

Added

• Document Editor - Split-view markdown editor with live preview
  • CodeMirror 6 with syntax highlighting and word wrap
  • Auto-save with 2s debounce
  • Keyboard shortcuts: ⌘B bold, ⌘I italic, ⌘K link, ⌘S save
  • Unsaved changes warning on navigation
• Document CRUD - Full create, read, update, delete via Web UI
  • PUT /api/docs/:id - Update document content
  • DELETE /api/docs/:id - Delete document and file
  • DocView edit/delete buttons with confirmation dialogs
• Collections Management - Add and remove collections from Web UI
  • AddCollectionDialog with folder path input
  • IndexingProgress component with real-time status
  • Re-index action per collection
• Keyboard Shortcuts - Global and editor shortcuts
  • ⌘N new note, ⌘K focus search, ⌘/ show help
  • ShortcutHelpModal with tactile keyboard key styling
  • Platform-aware modifier display (⌘ vs Ctrl)
• Quick Capture - Floating action button for instant note creation
  • CaptureModal with title, content, collection fields
  • Remembers last used collection
  • Shows indexing progress after creation

Changed

• Design Polish - Enhanced visual refinements
  • Hero Documents card with gradient and glow
  • Card hover lift effects with shadows
  • Enhanced glass effect with saturation
  • Search input with gradient focus border
  • Improved empty states with actionable suggestions
  • CaptureButton with pulse glow animation
• Fixed transparent backgrounds in dialogs, dropdowns, selects
• Changed collection remove icon from Trash to FolderMinus

0.7.0 - 2026-01-01

Added

• REST API write operations - Full CRUD for collections and documents
  • POST /api/collections - Add new collection
  • DELETE /api/collections/:name - Remove collection
  • POST /api/sync - Trigger re-index
  • POST /api/docs - Create document
  • POST /api/docs/:id/deactivate - Unindex document
  • GET /api/jobs/:id - Poll job status
• CSRF protection - Origin validation for mutating endpoints
• API token auth - X-GNO-Token header for non-browser clients

Changed

• Extracted collection CRUD to src/collection/ for API/CLI parity
• Config sync with mutex serialization (YAML → DB → memory)

0.6.1 - 2026-01-01

Added

• Auto-download models - Models download automatically on first use (CLI, MCP, Web UI)
• Offline mode - --offline flag and HF_HUB_OFFLINE/GNO_OFFLINE env vars
• Cross-process locking - Safe concurrent model downloads with stale lock recovery

0.6.0 - 2026-01-01

Added

• Tiered search modes - --fast (BM25 only, ~0.7s), default (with reranking, ~2-3s), --thorough (full pipeline with expansion, ~5-8s)
• Chunk-level reranking - Reranks best chunk per document (4K) instead of full documents, ~25× faster with same quality

Changed

• Default search now skips LLM query expansion (faster) but keeps reranking (quality)
• Refactored rerank pipeline with extracted helper functions for maintainability

Fixed

• Properly await store.close() in scripts
• Handle cleanupAndExit promises to prevent floating promise warnings

0.5.1 - 2026-01-01

Fixed

• Include fts5-snowball binaries in npm package

0.5.0 - 2026-01-01

Added

• Document-level BM25 - Full documents indexed (not chunks), finds terms across sections
• Snowball stemmer - FTS5 with multilingual stemming (20+ languages); “running” matches “run”
• Contextual chunking - Chunks embedded with document title prefix for context awareness
• Strong signal detection - Skip expensive LLM expansion when BM25 has confident match
• Tiered top-rank bonus - +0.05 for #1, +0.02 for #2-3 in RRF fusion
• Full-document reranking - Qwen3-Reranker scores complete documents (32K context)
• Full-document answers - Answer generation receives complete document content

Changed

• Original query now gets 2× weight in RRF fusion (prevents dilution by variants)
• Switched to Qwen3-Reranker-0.6B (Apache 2.0 license, 32K context)

0.4.0 - 2026-01-01

Added

• Web UI - Local web dashboard via gno serve
  • Dashboard with index stats and collection overview
  • Search page with BM25/hybrid search modes
  • Ask page with AI-powered answers and citations
  • Document viewer with syntax highlighting
  • Browse collections and documents
  • Model preset selector with hot-reload
  • Model download with progress polling
• REST API - Full HTTP API for programmatic access
  • /api/search - BM25 keyword search
  • /api/query - Hybrid search (BM25 + vector)
  • /api/ask - AI answers with citations
  • /api/docs, /api/doc - Document listing and retrieval
  • /api/presets - Model preset management
  • /api/models/pull - Model download with progress
• Convenience scripts - bun run serve and bun run serve:dev

Changed

• Extracted answer generation to shared src/pipeline/answer.ts module
• API and CLI now share identical pipeline code

0.3.5 - 2025-12-31

Changed

• README: Comprehensive rewrite with full MCP install commands, feature tables, doc links
• README: Added footer attribution

Fixed

• License badge spacing

0.3.4 - 2025-12-31

Fixed

• CI: Use Node.js 24 for npm OIDC trusted publishing support

0.3.3 - 2025-12-31

Fixed

• CI: Use npm for tarball install test (bun has issues with local tarballs)

0.3.2 - 2025-12-31

Fixed

• CI: Add TTY workarounds to publish workflow

0.3.1 - 2025-12-31

Fixed

• CI: Serial test execution on Windows fixes exit code issue
• CI: TTY workaround for macOS/Ubuntu test output
• CI: Use native Bun ecosystem for Dependabot

Changed

• Use npm trusted publishing (OIDC) instead of tokens

0.3.0 - 2025-12-31

Added

• 10 MCP installation targets (Claude Desktop, Claude Code, Cursor, Zed, Windsurf, etc.)
• LibreChat YAML config support
• Incremental sync (only new/modified files re-indexed)

0.2.1 - 2025-12-31

Added

• MCP Install CLI - One-command setup for AI assistant integration
  • gno mcp install - Configure gno as MCP server in Claude Desktop, Claude Code, or Codex
  • gno mcp uninstall - Remove gno from MCP configuration
  • gno mcp status - Check installation status across all targets
• Cross-platform support (macOS, Windows, Linux)
• Atomic config writes with automatic backup
• User and project scope support for Claude Code and Codex

0.2.0 - 2025-12-30

Added

• MCP server with stdio transport for AI assistant integration
• Tools: gno_search, gno_vsearch, gno_query, gno_get, gno_multi_get, gno_status
• Resources: gno://{collection}/{path} document access
• Plan for gno mcp install auto-configuration command
• Website: punchy tagline, 8 features in bento grid, theme-aware terminal glow

0.1.0 - 2025-12-30

Added

Core Features

• Hybrid Search Pipeline - BM25 + vector retrieval with RRF fusion
• HyDE Query Expansion - Hypothetical Document Embeddings for better semantic matching
• Cross-encoder Reranking - Two-stage retrieval with reranking for precision
• Multi-language Support - Automatic language detection with BCP-47 codes

CLI Commands

• gno init - Initialize a knowledge base with a collection
• gno collection add/list/remove/rename - Manage document collections
• gno update - Sync files from disk (no embedding)
• gno index - Full index with optional embedding
• gno embed - Generate vector embeddings
• gno search - BM25 keyword search
• gno vsearch - Vector semantic search
• gno query - Hybrid search with expansion and reranking
• gno ask - AI-powered Q&A with citations
• gno get - Retrieve single document by reference
• gno multi-get - Batch document retrieval
• gno ls - List indexed documents
• gno context add/list/rm - Manage retrieval context hints
• gno models list/use/pull/clear/path - Model management
• gno status - Index status and health
• gno doctor - Diagnostics and troubleshooting
• gno cleanup - Remove orphaned data
• gno skill install/uninstall/show/paths - Claude Code skill management

Document Processing

• Native Converters - Markdown, plain text, JSON, YAML, TOML, CSV
• External Converters - PDF (pdftotext), Office docs (pandoc), images (tesseract OCR)
• Smart Chunking - Semantic-aware document splitting
• File Walker - Configurable patterns, gitignore support, extension filters

Storage & Indexing

• SQLite Backend - Single-file database with FTS5 full-text search
• sqlite-vec - Cross-platform vector storage and similarity search
• Migrations - Schema versioning with automatic upgrades
• Mirror Cache - Converted document caching for fast re-indexing

AI/ML Integration

• Local Embeddings - GGUF models via node-llama-cpp (no API keys)
• Model Presets - slim (~1GB), balanced (~2GB), quality (~2.5GB)
• Reranker Models - Cross-encoder scoring for result quality
• LLM Abstraction - Pluggable providers (Anthropic, OpenAI, Ollama)

Developer Experience

• Output Formats - JSON, CSV, Markdown, XML, files protocol
• Verbose Mode - Detailed logging with --verbose
• Exit Codes - 0 (success), 1 (validation), 2 (runtime)
• Contract Tests - Schema validation for all outputs

Documentation

• Jekyll Website - Comprehensive docs at gno.sh
• VHS Terminal Demos - Animated CLI demonstrations
• Search Pipeline Guide - Deep dive into HyDE, RRF, reranking
• API Specifications - CLI spec, MCP spec, output schemas

Infrastructure

• Bun Runtime - Fast startup, native TypeScript
• Biome Linting - Zero-config via Ultracite preset
• GitHub Actions CI - Lint, typecheck, test matrix
• Beads Issue Tracking - Git-native dependency-aware tracking

Version History

View on GitHub