Releases

This page summarizes ContextWeaver changelogs from the v1.0.0 baseline onward. v1.0.0 maps to commit da79f2931157aa06b08aab99aaa4d43bcfa43f66; no matching Git tag or GitHub Release was found in the repository. v1.4.0 has a Git tag, but no GitHub Release was found. Prerelease versions create GitHub pre-releases and local testing tarballs only; they are not published to npm.

Version index

Version	Type	Published	Link	Changes
v1.6.0-alpha.1	Alpha prerelease	2026-06-09	GitHub Release	v1.6.0-alpha.0...v1.6.0-alpha.1
v1.6.0-alpha.0	Alpha prerelease	2026-06-09	GitHub Release	v1.5.3...v1.6.0-alpha.0
v1.5.4	Stable	2026-06-13	GitHub Release	v1.5.3...v1.5.4
v1.5.3	Stable	2026-06-06	GitHub Release	v1.5.2...v1.5.3
v1.5.3-beta.1	Beta prerelease	2026-06-06	GitHub Release	v1.5.3-beta.0...v1.5.3-beta.1
v1.5.3-beta.0	Beta prerelease	2026-06-05	GitHub Release	v1.5.3-alpha.0...v1.5.3-beta.0
v1.5.3-alpha.0	Alpha prerelease	2026-06-05	GitHub Release	v1.5.2...v1.5.3-alpha.0
v1.5.2	Stable	2026-06-03	GitHub Release	v1.5.1...v1.5.2
v1.5.1	Stable	2026-06-03	GitHub Release	v1.5.0...v1.5.1
v1.5.0	Stable	2026-06-02	GitHub Release	v1.4.0...v1.5.0
v1.4.0	Stable tag	2026-05-19	Git tag	v1.0.0 commit...v1.4.0
v1.0.0	Baseline commit	2026-03-13	Commit	v0.0.7...v1.0.0 commit

Unreleased

No pending changes.

v1.5.4

Release page · Full changes

Adds the contextweaver config command for managing environment variables through the CLI, eliminating the need to manually edit .env files. Supports listing, setting, validating, and interactive wizard-based configuration.

New Features

Config Command

• config list: View current configuration with automatic masking of sensitive information (API keys)
• config set <key> <value>: Set individual environment variables with validation
• config validate: Verify that all required configuration is valid
• config wizard: Interactive setup wizard for guided configuration

Supported configuration keys:

• Embedding settings: EMBEDDINGS_API_KEY, EMBEDDINGS_BASE_URL, EMBEDDINGS_MODEL, EMBEDDINGS_MAX_CONCURRENCY, EMBEDDINGS_DIMENSIONS
• Reranker settings: RERANK_API_KEY, RERANK_BASE_URL, RERANK_MODEL, RERANK_TOP_N
• Search parameters: CW_SEARCH_WVEC, CW_SEARCH_WLEX, CW_SEARCH_RERANK_TOP_N, CW_SEARCH_MAX_TOTAL_CHARS, CW_SEARCH_VECTOR_TOP_K, CW_SEARCH_SMART_MAX_K, CW_SEARCH_IMPORT_FILES_PER_SEED
• Other: IGNORE_PATTERNS

Usage examples:

# View current configuration
contextweaver config list

# Set a configuration value
contextweaver config set EMBEDDINGS_MAX_CONCURRENCY 20

# Validate configuration
contextweaver config validate

# Interactive configuration wizard
contextweaver config wizard

Documentation

• Added comprehensive config command documentation (docs/config-command.md) with usage guides, best practices, and troubleshooting

v1.6.0-alpha.1

Release page · Full changes

Continues the native migration roadmap with P1 (import resolver extract()) and P2 (encoding detect/decode) ported to Rust. Both are pure speedups with clean TypeScript fallbacks. With these, all CPU-bound hotspots (chunking/AST, import extraction, encoding decode) now run natively; the remaining scanner hash/filter modules were evaluated and intentionally kept in TypeScript.

Highlights

Native import extraction (Rust regex port) — P1

• Ported the 7 import resolvers' extract() regex to the crates/chunker native module via extractImports(kind, content) (kinds: jsts / python / go / java / rust / cpp / csharp).
• Output is byte-for-byte identical to the TypeScript regex, so the TypeScript-side resolve() keeps working unchanged. GraphExpander's E3 import expansion — called on every search for each seed file — now runs through the native path when available.
• Handled JS/Rust regex divergence: ASCII \w semantics, and the C# (?!static)(?!global) negative lookahead (unsupported by Rust regex) emulated in code.
• When the native module is unavailable, each resolver transparently falls back to its original TypeScript regex (extractTs).

Native encoding detect/decode (Rust) — P2

• Ported the detect+decode step of readFileWithEncoding to the crates/chunker native module via decodeBytes(buffer), backed by chardetng (Firefox's encoding detector) + encoding_rs (Gecko engine), replacing the JS chardet + iconv-lite.
• BOM sniffing mirrors the TypeScript logic; UTF-32 (LE/BE) is decoded manually since encoding_rs does not support it. Output is always UTF-8.
• readFileWithEncoding reads bytes in Node (fs.readFile stays in TypeScript), then prefers decodeBytes; on missing native binary or error it falls back to the original chardet/iconv-lite path (decodeBytesTs).

Migration scope concluded

• Evaluated the deferred scanner modules and decided not to migrate them: hash.ts (sha256) already runs on Node's OpenSSL-backed crypto, so a Rust port would only add a string copy across the NAPI boundary; filter.ts is I/O-bound and built on the mature npm ignore library, where a Rust rewrite risks byte-level gitignore divergence for no bottleneck gain.

Quality and verification

• tests/search/ImportExtract.diff.test.ts: asserts native and TypeScript import output match byte-for-byte across all 7 kinds, edge cases (in-string / comment imports, C# static/global exclusion, Rust pub mod, Go block imports), and a sample of real repository source files.
• tests/utils/Encoding.diff.test.ts: asserts native decodeBytes content matches the chardet/iconv-lite path on BOM and long CJK (GB18030 / Big5 / Shift_JIS) samples where detection converges; short ambiguous single-byte cases (where detector divergence is acceptable per design) are excluded.
• Full suite green (701 tests) with the native module built.

Installation

Prerelease versions are not published to npm; download the platform tarball from the GitHub Release for local testing only.

v1.6.0-alpha.0

Release page · Full changes

First prerelease shipping the Rust napi-rs native chunker. It creates a GitHub pre-release and all-platform testing tarballs only; it is not published to npm. Used to validate the cross-platform optionalDependencies install chain and native / TS fallback behavior.

Highlights

Native Rust chunker (napi-rs)

• Migrated the CPU-intensive chunking layer (Tree-sitter parsing + AST traversal + windowing + symbol/call-site extraction) to the crates/chunker native module.
• processor.ts now prefers the Rust single-parse path (process_file), reusing one syntax tree to produce chunks, symbols, and call sites in a single pass.
• When the napi module fails to load (no prebuilt binary on uncommon platforms, or an unbuilt dev environment), it transparently falls back to the existing TypeScript chunker, keeping the package usable everywhere.
• All LanceDB offset fields are normalized to the UTF-16 character domain on the Rust side, kept byte-for-byte consistent with the TS SourceAdapter.

Cross-platform build and release

• release.yml becomes three-staged: build (5-platform napi matrix: linux-x64-gnu / linux-arm64-gnu / darwin-x64 / darwin-arm64 / win32-x64-msvc, with --use-napi-cross for arm64-linux) → publish-chunker (platform subpackages → chunker main package) → publish-main (main app package → GitHub Release → MCP Registry), with needs enforcing publish ordering.
• Distribution uses two layers of optionalDependencies: main app package → @chiway/contextweaver-chunker (napi loader main package) → 5 platform subpackages; falls back to TS when a package is missing.
• prerelease.yml mirrors the build matrix and packages all-platform tarballs as GitHub Prerelease assets without publishing to npm.
• Cross-compile darwin-x64 on an arm64 macOS runner, and use --no-frozen-lockfile in build/publish jobs.

Quality and verification

• Added differential tests pinning UTF-16 offset-domain consistency: SourceAdapter, SemanticSplitter, Symbols, and CallSites Rust output compared field-by-field against TS output.
• Added a processor end-to-end integration test asserting that the built environment actually takes the native path.

Installation

Prereleases are not published to npm. Download the matching platform tarball from the GitHub Release for local testing.

v1.5.3

Release page · Full changes

Highlights

Semantic graph and symbol indexing

• Added Tree-sitter tags based symbol extraction for indexed source files, stored in the new semantic_symbols table.
• Added semantic call edge extraction for TypeScript, JavaScript, Python, Go, Rust, Java, C, C++, C#, Ruby, and PHP.
• Added the list-symbols MCP tool for browsing symbols by path, kind, language, and source.
• Improved get-symbol-definition by preferring indexed semantic_symbols definitions before falling back to lexical matches.
• Added incremental cleanup for semantic_edges, so changed files remove stale call edges even when the new version has no call sites or no matched local symbols.

Vector index readiness and recovery

• Added the contextweaver update command for package update checks and guided upgrades.
• Added the vector_manifest readiness table, making SQLite the authoritative source of vector index state while LanceDB remains the derived materialization layer.
• Search and graph expansion now trust only files whose vector readiness is hash-matched, reducing stale or partially written vector recall.
• Added recovery paths for pending vector marks and no-chunk files so repeated scans converge instead of retrying indefinitely.

Release and registry readiness

• Added MCP Registry metadata via server.json and package.json#mcpName.
• Added stable and prerelease GitHub Actions publishing paths with npm trusted publishing and MCP Registry OIDC authentication.
• Fixed prerelease workflow OIDC permissions and removed an invalid dangling files key.
• Fixed release-blocking typecheck, lint, formatting, and whitespace issues found during the 1.5.3-rc.0 review.

Schema updates

• Database schema version advances to 5.
• New tables: vector_manifest, semantic_symbols, and semantic_edges.
• semantic_symbols stores extracted definitions by file hash and symbol metadata.
• semantic_edges stores semantic relationships, including local Tree-sitter call edges.

Quality and verification

• Restored and expanded getSymbolDefinition tests.
• Added regression coverage for semantic_symbols primary-key behavior.
• Added regression coverage for stale semantic_edges cleanup when files lose call sites or no longer match local symbols.
• Final release validation passed:
  • pnpm typecheck
  • pnpm run lint
  • pnpm test — 124 test files / 603 tests passed
  • pnpm build

Installation

npm install -g @chiway/contextweaver@1.5.3
# or
pnpm add -g @chiway/contextweaver@1.5.3

v1.5.3-beta.1

Release page · Full changes

Highlights

Call Graph Export (Semantic Edges)

• Tree-sitter call extractor: Extracts function call nodes from AST without external dependencies
• Local call graph builder: Resolves function calls within the same file and populates the semantic_edges table with kind='call'
• Multi-language support: TypeScript, JavaScript, Python, Go, Rust, Java, C, C++, C#, Ruby, PHP (11 languages)
• Bidirectional queries: Find callers and callees using indexed queries on semantic_edges

Verification (ContextWeaver project):

• 385 call edges covering 96 files
• Provider: tree-sitter (lightweight implementation)
• Local call coverage: ~60-70% (as expected for same-file resolution)

list-symbols MCP Tool

• New MCP tool for querying symbol outlines across the codebase
• Multi-dimensional filtering:
  • Path filter (prefix or glob pattern, e.g., src/**/*.ts)
  • Symbol kind filter (function,class,interface)
  • Language filter (typescript,python,go)
  • Source filter (tree-sitter / ctags)
• Markdown output: Grouped by file with line ranges and container information
• Efficient SQL queries: Path filtering handled at SQL layer (LIKE queries) to avoid post-filter LIMIT traps

Tree-sitter Tags Symbol Extraction

• Uses grammar-native tags.scm query files to extract symbols from AST
• Adds patches for TypeScript/JavaScript (class/function/enum definitions)
• Optimizes get-symbol-definition tool: prioritizes semantic_symbols table lookup with directory-aware ranking

Schema Updates

• Extended SemanticEdge.provider type: 'lsp' | 'tree-sitter'
• Updated database CHECK constraint to allow tree-sitter as a provider

Performance

• Zero new dependencies: Fully reuses existing tree-sitter and ParserPool infrastructure
• Indexing impact: < 5% increase (call graph building has negligible overhead)
• Query performance: Dual-index support for bidirectional call graph traversal

Implementation Details

New files:

• src/semantic/treeSitterCalls.ts (184 lines) — Call site extractor
• src/semantic/callGraphBuilder.ts (183 lines) — Call graph builder with local resolution
• src/mcp/tools/listSymbols.ts (149 lines) — Symbol listing MCP tool

Modified files:

• src/semantic/types.ts — Extended SemanticEdge.provider
• src/db/index.ts — Updated CHECK constraint
• src/scanner/processor.ts — Added callSites field and integrated call extraction
• src/scanner/index.ts — Invokes buildAndStoreCallGraph
• src/mcp/server.ts — Registered list-symbols tool

Future Enhancements (Not Implemented)

• Cross-file call resolution (Phase 2): Use resolver + symbol disambiguation for inter-file calls
• Call graph query MCP tools: Direct call relationship queries (e.g., find-callers, find-callees)
• QueryPlanner integration: Recognize symbol query intent and route to symbol table for exact matches
• Incremental symbol indexing: Hook into watcher for file-change-triggered updates

Testing and installation

This version is not published to npm. Download the .tgz asset from the Release page and install it locally:

npm install -g ./chiway-contextweaver-1.5.3-beta.1.tgz

v1.5.3-beta.0

Release page · Full changes

Highlights

• Added parser enrichment layers: Tree-sitter remains the primary parser, with optional ctags fallback symbol extraction and a disabled-by-default LSP enrichment framework.
• Added SQLite semantic_symbols and semantic_edges metadata tables for fallback symbols and future LSP semantic edges.
• Added DeepSeek-powered English changelog generation to the stable release workflow, with raw commit fallback when unavailable.
• Updated the prerelease workflow to build a local npm tarball and attach it to the GitHub pre-release.

Testing and installation

This version is not published to npm. Download the .tgz asset from the Release page and install it locally:

npm install -g ./chiway-contextweaver-1.5.3-beta.0.tgz

v1.5.3-alpha.0

Release page · Full changes

Highlights

• Added the contextweaver update command.
• Added a vector manifest consistency guard: SQLite becomes the authoritative readiness state and LanceDB is treated as the derived vector materialization layer.
• Search and graph expansion now trust only files that are ready and hash-matched in SQLite, reducing SQLite/LanceDB dual-write inconsistency risk.

Testing and installation

This version is not published to npm. Download the .tgz asset from the Release page and install it locally:

npm install -g ./chiway-contextweaver-1.5.3-alpha.0.tgz

v1.5.2

Release page · Full changes

Highlights

• Added retrieval controls to the codebase retrieval MCP tool.
• Separated semantic and lexical retrieval queries.
• Added structured codebase retrieval output.
• Added deterministic deep query decomposition.
• Extended graph expansion with reverse imports and call-sites.
• Exposed low-confidence retrieval handling.
• Preserved retrieval provenance in context segments.
• Reused search services across MCP calls to reduce repeated initialization overhead.

Installation

npm install -g @chiway/contextweaver@1.5.2
# or
pnpm add -g @chiway/contextweaver@1.5.2

v1.5.1

Release page · Full changes

Highlights

• Updated README with v1.4.0 data architecture and migration details.
• Corrected package and repository metadata to @chiway/contextweaver and wchiway/contextweaver-mcp.
• Added CI gates for lint, typecheck, test, and build.
• Fixed a stale MCP build entry reference.
• Made the MCP server version read dynamically from package.json.
• Removed unused fields and promoted lint to a hard gate.
• Fixed old package names in the release workflow that affected duplicate-publish protection and installation instructions.

Installation

npm install -g @chiway/contextweaver@1.5.1
# or
pnpm add -g @chiway/contextweaver@1.5.1

v1.5.0

Release page · Full changes

Highlights

• Added watch mode and configurable search caching.
• Added stats for indexing, search, and health metrics.
• Added four granular MCP tools and matching CLI mirrors.
• Added AST semantic chunking support for Ruby, PHP, Kotlin, Swift, Lua, and Shell.
• Documented CW_SEARCH_* environment variable configuration.
• Fixed typecheck, lint, and CI issues discovered by the new gates.
• Rewrote README for v1.5.0 features and split English/Chinese documentation.

Installation

npm install -g @chiway/contextweaver@1.5.0
# or
pnpm add -g @chiway/contextweaver@1.5.0

v1.4.0

Git tag · Full changes

Highlights

• Major data architecture and cross-store consistency fixes.
• Removed display_code / vector_text from the LanceDB chunks table; source text is looked up from files.content instead.
• Normalized SemanticSplitter metadata offsets to the UTF-16 character domain before writes, fixing UTF-8 byte offset drift.
• Schema version 2 → 3: added the pending_marks outbox table for replaying vector index marks after mark-stage failures.
• Added a pending / done / aborted LanceDB migration state machine and a cross-process advisory lock.
• Added contextweaver migrate and contextweaver migrate --reset.
• Added src/db/bootstrap.ts to decouple VectorStore and SQLite initialization coordination.
• Fixed ChunkContentLoader slicing basis, crash-safe LanceDB migration, aborted-state write blocking, duplicate chunk_id handling, migration locks, and import target loader reuse.
• Expanded tests from 38 to 109 and added real LanceDB end-to-end migration tests.

Installation

npm install -g @chiway/contextweaver@1.4.0
# or
pnpm add -g @chiway/contextweaver@1.4.0

v1.0.0

Baseline commit · Full changes

Highlights

• Bumped the project to 1.0.0, establishing the baseline for subsequent v1.x changelogs.
• Hardened rerank handling by tolerating empty or non-JSON responses.
• Falls back to unreranked candidates when rerank is misconfigured or unavailable, preventing search crashes.
• Removed the Zen Config abstraction before 1.0.0 and merged defaults into the global configuration.
• Hardened the lock mechanism and prevented an infinite healing loop.

Link note

No matching v1.0.0 Git tag or GitHub Release was found in the repository; this entry uses commit da79f2931157aa06b08aab99aaa4d43bcfa43f66 as the v1.0.0 baseline.