Changelog
What changed in each version.
remimem follows Semantic Versioning. While the version is below 1.0.0, a minor version may contain breaking changes. The changelog in the repository is the source; this page mirrors it.
0.1.0 (2026-11-15)
First release.
- Indexing: a scanner that respects
.gitignoreand skips vendored, generated, binary, oversized, and secret-looking files; tree-sitter parsing for TypeScript, TSX, JavaScript, Python, and Go; one chunk per function, method, type, and class header; line windows for every other text file. - Storage: one SQLite file per project, with an identifier-aware keyword index, vectors in
sqlite-vec, and numbered migrations. - Embeddings: a local model by default (
Xenova/all-MiniLM-L6-v2), optional OpenAI and Voyage providers, and a cache so unchanged text is never embedded twice. - Search: keyword and vector rankings fused by reciprocal rank, exact symbol names first, a reason on every result.
- Call graph: links with a confidence (
exact,import,name-match), used to add the neighbors of top results as signatures. - Packing: answers fitted into a token budget, grouped by file in line order, never over the budget.
- Project memory: decisions, conventions, gotchas, and todos linked to code, mirrored to
.remimem/memory.json, marked stale when the code changes. - MCP server: nine tools and the
remimem://overviewresource, with changed files re-indexed before each call. - CLI:
init,index,watch,serve,search,context,memory,stats,doctor, andbench. - Benchmark: three open-source repositories, a grep-and-read baseline, ablations, and a held-out test split.
The project was developed under the working name codemem and renamed before release, because that name was taken on npm.