v0.4.0

Lore v0.4.0

Full migration to SCIP-first indexing architecture. Tree-sitter is completely removed.

Highlights

• SCIP+LSP architecture — Single-pass SCIP indexing for 23 languages, LSP enrichment overlay
• C symbol directionality fix — Header forward declarations correctly resolve to implementations
• Macro name extraction — Preprocessor macros indexed with real #define names via SCIP occurrence ranges
• Test coverage at 92% — 1,664 tests, suite rewritten from scratch

Breaking Changes

• Tree-sitter dependencies removed — no native grammar builds required
• Minimum Node.js 22 (unchanged)

Features

• SCIP+LSP architecture replaces tree-sitter extraction (#458)
• Parent symbol tracking from SCIP descriptor chains (#249)
• Virtual dispatch edge materialization (#221)

Fixes

• Forward declaration → definition directionality for C headers (#460)
• Preprocessor macro name recovery from SCIP ranges (#460)
• SCIP backtick-quoted identifiers handled as atomic names (#460)
• Call-graph resolution scoped by layer (#397)
• TOML array parsing handles quoted commas (#446)
• 20+ additional bug fixes

Refactoring

• CLI, DB, schema, indexer, and enrichment modules split into focused units (#438–#444)

Full Changelog

v0.3.9...v0.4.0

v0.3.9

v0.3.9

Features

• Virtual dispatch materialization — lore_graph now materializes virtual dispatch edges, enabling polymorphic call resolution through the call graph (#221)
• Automatic transitive traversal — lore_graph and lore_dependents now automatically follow edges up to 5 hops; user-facing depth and limit parameters removed in favor of sensible internal defaults (#221)
• Truncation flag — lore_graph and lore_dependents results include a truncated field when results hit the internal limit

Fixes

• Virtual dispatch follow-up — corrected edge cases in SCIP virtual dispatch resolution (#222)
• Tree-sitter end_line — fixed end-line calculation in tree-sitter source indexing (#219)

Benchmarks & Tests

• Expanded Postgres benchmark question set with corrected expected answers (#217, #218)
• Call-graph focused benchmark suite (#219)
• Structured benchmark report output (#223)

Docs

• Updated lore_graph and lore_dependents tool descriptions in README and architecture docs to reflect automatic transitive traversal and virtual dispatch materialization

Full changelog: v0.3.8...v0.3.9

v0.3.8

v0.3.8

Features

• Incremental indexing — baseline + overlay architecture for fast file-level re-indexing without full rebuilds (#213)
• SCIP auto-install — auto-install SCIP indexers and generate compile_commands.json for C/C++ (#211)
• Auto-generate tsconfig — broad tsconfig generation for TypeScript SCIP indexing (#205)
• New MCP tools — lore_trace (execution path tracing), lore_diff (API surface diffing), lore_cohesion (module cohesion metrics), lore_dependents (unified reverse-dependency queries) (#198, #201, #202, #204)
• Auto-index on MCP start — mcp --root now indexes automatically (#182)
• Benchmark overhaul — Gson + Postgres repos, per-test coverage infra, ground truth updates (#206)

Refactors

• Pre-release consolidation — removed ~850 lines of dead code, backward-compat shims, and code duplication (#215):
  • createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - cray - creat(DDL - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpng - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() ark - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - createLoreth - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - createLoreMcpServer() - cray - creat(DDL - createLoreMcpServer() - createLoreMcpServer() - createLoacO - createLoreMcpServer() - createLoreMcpServer() - `createLoreMcpServer)
• • • • • • • • • • -path storage (#20- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -path storage eferences; added new tools to tool table and mermaid diagram
• Added CLAUDE.md project instructions (#214)
• Added docs/incremental-index-design.md and docs/complex-codebase-questions.md
• README refresh with current tools and architecture (#207)

Full changelog: v0.3.7...v0.3.8

v0.3.7

v0.3.7

Features

• Benchmark suite overhaul — overhauled copilot benchmark suite with indexing performance improvements, full prompts in results, and improved scorer arrow-line matching (#181, #183)

Refactors

• Removed low-value MCP tools — removed lore_annotations, lore_coverage, lore_writeback, lore_analyze, lore_architecture, and module edge kind from lore_graph, reducing tool surface for cleaner agent workflows (#186)

Fixes

• Improved benchmark scorer arrow-line matching and resolved dynamic import handling (#183)

Docs

• Updated README benchmark numbers (8.8× faster, 97% fewer tokens, +33pp correctness)
• Improved README mermaid diagram layout and fixed benchmark-results formatting (#185)
• Removed stale tool references from README.md and docs/architecture.md tool tables and mermaid diagrams

Full changelog: v0.3.6...v0.3.7

v0.3.6

v0.3.6

Features

• SCIP-primary architecture — SCIP indexers are now the default source stage for all languages that have one; tree-sitter is the fallback. New src/scip/ module directory with config, enrichment coordinator, index reader, registry, and protobuf definitions. New ScipEnrichmentStage pipeline stage writes definition locations and type metadata from SCIP data before LSP enrichment. (#163)
• Benchmark library — new benchmark test suite for evaluating Lore agent performance against ground-truth answers (#176)

Fixes

• symbol_relationships.line is now nullable for missing definition locations (#177)

Refactors

• Domain directory restructure — src/ reorganized from flat indexer/ into domain directories: parsing/, discovery/, git/, lsp/, db/, docs/, embeddings/, resolution/, server/, testing/, scip/ (#175)
• Extractor condi- Extractor condi- Extractor condi- Extractor condi- ExtractorCo- Extractor cond, SCIP default testing,- Extractor condi-

#####ocs

• Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Upd- Updgelog:** v0.3.5...v0.3.6

v0.3.5

v0.3.1 — Correctness fixes

Bug fixes (6)

Critical

• cli.ts: Fix wrong table names in MCP startup stats (call_graph → symbol_refs, documentation → docs) — totalEdges and totalDocs previously always reported 0

High

• source-index.ts: NULL out cross-file symbol_refs.callee_id, type_refs.type_id, and symbol_relationships.target_symbol_id before deleting a file in update mode — prevents FK constraint violation that aborted the entire update transaction

Medium

• source-index.ts: Replace single wrapping transaction with batched transactions (200 files/batch) for crash-resilient checkpointing

Low

• import-resolution.ts, lsp-enrichment.ts, dependency-api.ts: Wrap bulk writes in transactions (~100x faster for large codebases)
• embedding.ts: Chunk symbol embedding deletes into batches of 900 to stay within SQLite SQLITE_MAX_VARIABLE_NUMBER limit

Tests

• 4 new regression tests covering all fixes (1,157 total tests pass)

v0.3.4

v0.3.4

Features

• Native Node.js embeddings (#158) — Replaced the Python subprocess embedding system (SentenceTransformersProvider + torch + sentence-transformers) with TransformersJsProvider using @huggingface/transformers. Embedding models now run natively in Node.js via ONNX Runtime — no Python, pip, or torch required. Lore is now fully self-contained on Node >= 22.

• New default embedding model — Changed from Qwen/Qwen3-Embedding-4B (4B params, GPU + Python required) to nomic-ai/nomic-embed-text-v1.5 (0.1B params, 768-dim, CPU-friendly, ONNX-ready, Matryoshka dimension support).

Breaking Changes

• SentenceTransformersProvider removed from public API → use TransformersJsProvider
• Qwen3EmbeddingProvider factory removed
• ensurePythonDeps removed (no longer needed)
• Default embedding model changed to nomic-ai/nomic-embed-text-v1.5

Docs

• Updated README.md embeddings section (Python → Transformers.js)
• Updated docs/architecture.md embedder module description and pipeline diagram

Internal

• Removed src/indexer/ensure-python-deps.ts and its tests
• Added @huggingface/transformers@^3.8.1 dependency

Full Changelog: v0.3.3...v0.3.4

v0.3.3

v0.3.3

Features

• Reverse/inbound graph queries (#151) — lore_graph now accepts target_id for inbound edge queries across all five graph kinds (call, import, module, inheritance, type_dependency). Enables "who calls X?" and "what depends on X?" questions that were previously impossible without dumping all edges.

Fixes

• 18 extractor bug fixes + grammar hardening (#148) — Comprehensive parsing audit across all extractors: fixed SYMBOL_NODE_TYPES completeness in 12 extractors, added missing C++ extraction patterns (enums, typedefs, forward declarations), eliminated empty-name symbol leaks in 6 extractors, added pipeline-level empty-name guard. Swapped incompatible grammars for Lua, Zig, and Elm. Added ESM-only grammar fallback and Zig dual-grammar support.
• Child process cleanup on SIGINT/SIGTERM (#156) — New process-tracker module ensures Python embedding s- Child process cleanup on SIGINT/SIGTERM (#156) — New is- **Child procesha- **Child process cyn- **Child process ore - **Child psec- **Child process cleanup on SIGINT/SIGTERM** (#156) — New process-trackermodule ensures Python embedding s- **Child process cleanup on SIGINT/SIGTERM** (#156) — Newis- **Child procesha- **Child process cyn- **Child process ore - **Child psec- Child process cleanup on SIGINT/SIGTERM (#156) �dcod- **Chilon references).

Internal

• Dart support removed (#148, #150) — No compatible tree-sitter grammar exists; language count 24 → 23.
• LSP smoke test coverage (#150) — Real-server smoke tests for TypeScript and Pyright LSPs.
• Test hardening (#148) — Removed all test.skipIf(!result) patterns; all 1,385 tests pass with 0 skipped.

Full Changelog: v0.3.2...v0.3.3

v0.3.2

What's Changed

• fix(indexer): restore full C symbol extraction — 5 bug fixes by @jafreck in #146
• chore: bump version to 0.3.2 by @jafreck in #147

Full Changelog: v0.3.1...v0.3.2

v0.3.1

v0.3.1 — Correctness fixes

Bug fixes (6)

Critical

• cli.ts: Fix wrong table names in MCP startup stats (call_graph → symbol_refs, documentation → docs) — totalEdges and totalDocs previously always reported 0

High

• source-index.ts: NULL out cross-file symbol_refs.callee_id, type_refs.type_id, and symbol_relationships.target_symbol_id before deleting a file in update mode — prevents FK constraint violation that aborted the entire update transaction

Medium

• source-index.ts: Replace single wrapping transaction with batched transactions (200 files/batch) for crash-resilient checkpointing

Low

• import-resolution.ts, lsp-enrichment.ts, dependency-api.ts: Wrap bulk writes in transactions (~100x faster for large codebases)
• embedding.ts: Chunk symbol embedding deletes into batches of 900 to stay within SQLite SQLITE_MAX_VARIABLE_NUMBER limit

Tests

• 4 new regression tests covering all fixes (1,157 total tests pass)