Mempalace Ts

MemPalace.ts

Local AI memory system for Claude Code, ChatGPT, and any MCP-compatible tool.

TypeScript rewrite of MemPalace (87102fb). Native ONNX Runtime for embeddings — ~2× faster mining on Apple Silicon than the Python reference on the same corpus.

[![][version-shield]][release-link] [![][node-shield]][node-link] [![][license-shield]][license-link]

---

Why TypeScript

| | Python | TypeScript | |---|---|---| | Vector DB | ChromaDB (~500ms startup) | sqlite-vec (~15ms startup) | | Embeddings runtime | PyTorch CPU | ONNX Runtime (Apple Silicon native) | | Mining speed | baseline | ~2× faster (measured on identical corpus) | | Search | Semantic + closet boost + hybrid keyword | Same (parity) | | Cross-lingual | 100+ languages via e5-small | Same (parity) | | LLM rerank | Haiku only | Any OpenAI-compatible API | | Localization | 13 locales | 9 locales (EN, RU, DE, FR, ES, IT, PT-BR, HI, ID) | | Dependencies | Python + chromadb + sentence-transformers | Node.js only | | Tests | 92 | 663 |

Quick Start

npm install @vlaushkin/mempalace-ts

mempalace init ~/projects/myapp
mempalace mine ~/projects/myapp
mempalace search "why did we switch to GraphQL"

MCP Server

claude mcp add mempalace -- node dist/mcp/server.js

26 tools: mempalace_search, mempalace_add_drawer, mempalace_kg_query, mempalace_diary_write, mempalace_create_tunnel, mempalace_get_drawer, mempalace_ping, and [19 more](src/mcp/tools.ts).

Multilingual Search

{ "multilingual": true }

Dual-index architecture: English queries use all-MiniLM-L6-v2, non-English queries auto-route to multilingual-e5-small (100+ languages). Both indexes updated in one transaction.

EN "machine learning"    RU "машинное обучение"    -> 0.95
EN "git hooks setup"     DE "git hooks einrichten" -> 0.91

All Commands

mempalace init                     # detect rooms, create config
mempalace mine                     # mine project files (closets + halls auto-tagged)
mempalace mine  --mode convos      # mine conversation exports
mempalace search "query"                # hybrid semantic + closet-boosted search
mempalace search "query" --rerank       # + LLM rerank (any OpenAI-compatible API)
mempalace sweep               # message-granular safety net for Claude Code JSONL
mempalace diary-ingest             # ingest YYYY-MM-DD.md daily summaries
mempalace export --format jsonl         # dump palace (json/jsonl/markdown)
mempalace wake-up                       # load L0 + L1 context
mempalace compress --wing myapp         # AAAK compression
mempalace status                        # palace overview
mempalace migrate export/import/verify  # ChromaDB -> sqlite-vec

Development

npm install
npm run build          # tsup -> dist/
npm run test           # vitest (663 tests)
npm run lint           # biome check
npm run typecheck      # tsc --noEmit

Benchmarks

On the 500-question LongMemEval retrieval benchmark, the default search pipeline achieves R@5 = 98.0%, NDCG@5 = 0.932. On cross-lingual retrieval (Russian queries, English documents), multilingual-e5-small reaches R@5 = 89.0%. Reproduction instructions in [benchmarks/](benchmarks/).

License

MIT -- see [LICENSE](LICENSE).

[version-shield]: https://img.shields.io/badge/version-1.2.0-4dc9f6?style=flat-square&labelColor=0a0e14 [release-link]: https://github.com/vlaushkin/mempalace.ts/releases [node-shield]: https://img.shields.io/badge/node-22+-7dd8f8?style=flat-square&labelColor=0a0e14&logo=node.js&logoColor=7dd8f8 [node-link]: https://nodejs.org/ [license-shield]: https://img.shields.io/badge/license-MIT-b0e8ff?style=flat-square&labelColor=0a0e14 [license-link]: https://github.com/vlaushkin/mempalace.ts/blob/main/LICENSE

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: vlaushkin
• Source: vlaushkin/mempalace.ts
• License: MIT

Install and usage instructions live in the source repository linked above.