vectorgrep v0.3.0

vectorgrep now holds up when several agents use it at the same time, and tells
agents itself when and how to use it.

Added

• Usage instructions for agents: the server sends instructions, and every
  tool description says when to use it — semantic search for questions about
  what code does, grep for exact text, init once per project, index_update
  for routine refreshes. These reach every agent, including Explore and Plan
  subagents, which don't receive a project's CLAUDE.md (#53).
• search_symbols can filter by enum and module (#41).

Changed

• Searching a project that was never indexed now answers "No index found …
  Run 'init'" instead of an empty result, and no longer creates empty tables
  (#52).
• Requests to Ollama and OpenAI time out after 60 seconds (Ollama's
  availability check after 5 seconds) and are retried, instead of waiting
  forever for a server that doesn't answer (#33).

Fixed

• Parallel tool calls, e.g. from subagents sharing one server: concurrent
  index_update calls added the same new files twice or failed with LanceDB
  commit conflicts, and a search started during init or reindex kept
  returning nothing afterwards. Index writes for a project now run one after
  another, and searches wait for them (#52).
• index_update indexed every file a second time when it couldn't read the
  existing file hashes (#32).
• index_status showed wrong file, chunk and symbol counts after
  index_update (#37).
• Projects using different transformers.js models in the same server process
  overwrote each other's model (#30).

Internal

• Test coverage is measured in CI and may not drop below a floor (currently
  84 % statements, 71 % branches, 90 % functions, 85 % lines). New tests cover
  input sanitizing, retries, the embedding cache, result formatting and all
  tool handlers (#54, #55).
• The README installs from npm instead of cloning (#51); CLAUDE.md now holds
  project guidance only, open work is tracked in GitHub issues (#67).

Upgrade notes

• No reindex required — the index format is unchanged since 0.2.0.
• Wrong counts in index_status from earlier versions are corrected the next
  time index_update finds changes, or right away with reindex.

vectorgrep v0.2.0

The first public release on npm, under the new name vectorgrep — plus a long
list of search and indexing fixes found in a code audit and in an end-to-end
test against a real ~2,700-file project.

Added

• Published on npm as vectorgrep: npm install -g vectorgrep or
  npx -y vectorgrep, no clone and build needed. Releases are published with
  npm trusted publishing and carry a provenance attestation.

Changed

• Renamed the project to vectorgrep (npm package, CLI command, MCP server
  name and GitHub repository, formerly codebase-semantic-search). The storage
  location ~/.vectordb/ and the .vectordb.json config file are unchanged.
• Node.js 20 or newer is required (Node 18 is end-of-life).
• embedding.model is optional; each embedding provider uses its own default
  model when none is configured.
• Include/exclude patterns are matched with minimatch in both discovery modes.
  Files in dot-directories such as .github/ are no longer indexed in git mode
  unless an include pattern names them.

Fixed

• LanceDB filters on camelCase columns never matched, because double-quoted
  identifiers are string literals: index_update duplicated modified files and
  kept deleted ones, filePattern and symbolTypes filters returned nothing, and
  exact symbol-name matching never ran (#24).
• Reused chunk vectors were stored as NaN during incremental updates (#24).
• The default config was a shared object mutated across projects and calls, so
  settings from one init leaked into other projects (#25).
• The transformers.js fallback persisted the wrong embedding model, breaking
  every search and index_update after init (#18, #25).
• LineChunker looped forever when overlapLines >= maxChunkLines (#26).
• AST chunking dropped the bodies of long functions and all code outside
  declarations, and produced duplicate chunk ids for symbols sharing a line
  range (#26).
• Tree-sitter WASM lookup used require.resolve, which doesn't exist in ESM (#26).
• In git mode, default exclude patterns never matched, include was ignored,
  non-ASCII filenames were skipped, and a failing git ls-files produced an empty
  index instead of falling back to glob (#27).
• init and reindex silently discarded the entire index while reporting success
  when another MCP server process recreated the tables during indexing; searches
  then returned nothing (#47).

Internal

• Community health files (CONTRIBUTING.md, CODE_OF_CONDUCT.md, SECURITY.md,
  issue/PR templates), CI on Node 20/22 and Dependabot.
• Release process: a PR labeled release:major|minor|patch publishes to npm
  on merge, tags the version and uses this changelog section as the GitHub
  release notes. See RELEASING.md.
• Removed the unused dependencies @anthropic-ai/sdk and chokidar; bumped
  dependencies to resolve security advisories, with sharp and adm-zip pinned
  via overrides.
• Upgraded vitest to v4, zod to v4, @huggingface/transformers to v4 and
  glob to v13.
• CI and release skip onnxruntime-node's CUDA download, which repeatedly timed
  out (#48).

Upgrade notes

• Run reindex once after upgrading. Existing indexes may contain duplicate or
  stale chunks, miss code that was previously dropped, or record the wrong
  embedding model.