[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-678":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":23,"hasPages":25,"topics":26,"createdAt":10,"pushedAt":10,"updatedAt":33,"readmeContent":34,"aiSummary":35,"trendingCount":16,"starSnapshotCount":16,"syncStatus":36,"lastSyncTime":37,"discoverSource":38},678,"mempalace","MemPalace\u002Fmempalace","MemPalace","The best-benchmarked open-source AI memory system. And it's free.","http:\u002F\u002Fmempalaceofficial.com\u002F",null,"Python",55370,7191,318,261,0,105,1874,3397,775,45,"MIT License",false,"develop",true,[27,28,29,30,31,32],"ai","chromadb","llm","mcp","memory","python","2026-06-12 02:00:17","> [!CAUTION]\n> **Scam alert.** The only official sources for MemPalace are this\n> [GitHub repository](https:\u002F\u002Fgithub.com\u002FMemPalace\u002Fmempalace), the\n> [PyPI package](https:\u002F\u002Fpypi.org\u002Fproject\u002Fmempalace\u002F), and the docs site at\n> **[mempalaceofficial.com](https:\u002F\u002Fmempalaceofficial.com)**. Any other\n> domain — including `mempalace.tech` — is an impostor and may distribute\n> malware. Details and timeline: [docs\u002FHISTORY.md](docs\u002FHISTORY.md).\n\n> [!IMPORTANT]\n> **🚨 Claude Code sessions expire in 30 days w\u002Fout auto-save hooks wired!** **[Read this →](https:\u002F\u002Fgithub.com\u002FMemPalace\u002Fmempalace\u002Fdiscussions\u002F1388)**\n\n\n\u003Cdiv align=\"center\">\n\n\u003Cimg src=\"assets\u002Fmempalace_logo.png\" alt=\"MemPalace\" width=\"240\">\n\n# MemPalace\n\nLocal-first AI memory. Verbatim storage, pluggable backend, 96.6% R@5 raw on LongMemEval — zero API calls.\n\n[![][version-shield]][release-link]\n[![][python-shield]][python-link]\n[![][license-shield]][license-link]\n[![][discord-shield]][discord-link]\n\n\u003C\u002Fdiv>\n\n---\n\n## What it is\n\nMemPalace stores your conversation history as verbatim text and retrieves\nit with semantic search. It does not summarize, extract, or paraphrase.\nThe index is structured — people and projects become *wings*, topics\nbecome *rooms*, and original content lives in *drawers* — so searches\ncan be scoped rather than run against a flat corpus.\n\nThe retrieval layer is pluggable. The current default is ChromaDB; the\ninterface is defined in [`mempalace\u002Fbackends\u002Fbase.py`](mempalace\u002Fbackends\u002Fbase.py)\nand alternative backends can be dropped in without touching the rest of\nthe system.\n\nNothing leaves your machine unless you opt in.\n\nArchitecture, concepts, and mining flows:\n[mempalaceofficial.com\u002Fconcepts\u002Fthe-palace](https:\u002F\u002Fmempalaceofficial.com\u002Fconcepts\u002Fthe-palace.html).\n\n---\n\n## Install\n\n```bash\npip install mempalace\nmempalace init ~\u002Fprojects\u002Fmyapp\n```\n\n## Quickstart\n\n```bash\n# Mine content into the palace\nmempalace mine ~\u002Fprojects\u002Fmyapp                    # project files\nmempalace mine ~\u002F.claude\u002Fprojects\u002F --mode convos   # Claude Code sessions (scope with --wing per project)\n\n# Search\nmempalace search \"why did we switch to GraphQL\"\n\n# Load context for a new session\nmempalace wake-up\n```\n\nFor Claude Code, Gemini CLI, MCP-compatible tools, and local models, see\n[mempalaceofficial.com\u002Fguide\u002Fgetting-started](https:\u002F\u002Fmempalaceofficial.com\u002Fguide\u002Fgetting-started.html).\n\n---\n\n## Benchmarks\n\nAll numbers below are reproducible from this repository with the commands\nin [`benchmarks\u002FBENCHMARKS.md`](benchmarks\u002FBENCHMARKS.md). Full\nper-question result files are committed under `benchmarks\u002Fresults_*`.\n\n**LongMemEval — retrieval recall (R@5, 500 questions):**\n\n| Mode | R@5 | LLM required |\n|---|---|---|\n| Raw (semantic search, no heuristics, no LLM) | **96.6%** | None |\n| Hybrid v4, held-out 450q (tuned on 50 dev, not seen during training) | **98.4%** | None |\n| Hybrid v4 + LLM rerank (full 500) | ≥99% | Any capable model |\n\nThe raw 96.6% requires no API key, no cloud, and no LLM at any stage. The\nhybrid pipeline adds keyword boosting, temporal-proximity boosting, and\npreference-pattern extraction; the held-out 98.4% is the honest\ngeneralisable figure.\n\nThe rerank pipeline promotes the best candidate out of the top-20\nretrieved sessions using an LLM reader. It works with any reasonably\ncapable model — we have reproduced it with Claude Haiku, Claude Sonnet,\nand minimax-m2.7 via Ollama Cloud (no Anthropic dependency). The gap\nbetween raw and reranked is model-agnostic; we do not headline a \"100%\"\nnumber because the last 0.6% was reached by inspecting specific wrong\nanswers, which `benchmarks\u002FBENCHMARKS.md` flags as teaching to the test.\n\n**Other benchmarks (full results in [`benchmarks\u002FBENCHMARKS.md`](benchmarks\u002FBENCHMARKS.md)):**\n\n| Benchmark | Metric | Score | Notes |\n|---|---|---|---|\n| LoCoMo (session, top-10, no rerank) | R@10 | 60.3% | 1,986 questions |\n| LoCoMo (hybrid v5, top-10, no rerank) | R@10 | 88.9% | Same set |\n| ConvoMem (all categories, 250 items) | Avg recall | 92.9% | 50 per category |\n| MemBench (ACL 2025, 8,500 items) | R@5 | 80.3% | All categories |\n\nWe deliberately do not include a side-by-side comparison against Mem0,\nMastra, Hindsight, Supermemory, or Zep. Those projects publish different\nmetrics on different splits, and placing retrieval recall next to\nend-to-end QA accuracy is not an honest comparison. See each project's\nown research page for their published numbers.\n\n**Reproducing every result:**\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FMemPalace\u002Fmempalace.git\ncd mempalace\npip install -e \".[dev]\"\n# see benchmarks\u002FREADME.md for dataset download commands\npython benchmarks\u002Flongmemeval_bench.py \u002Fpath\u002Fto\u002Flongmemeval_s_cleaned.json\n```\n\n---\n\n## Knowledge graph\n\nMemPalace includes a temporal entity-relationship graph with validity\nwindows — add, query, invalidate, timeline — backed by local SQLite.\nUsage and tool reference:\n[mempalaceofficial.com\u002Fconcepts\u002Fknowledge-graph](https:\u002F\u002Fmempalaceofficial.com\u002Fconcepts\u002Fknowledge-graph.html).\n\n## MCP server\n\n29 MCP tools cover palace reads\u002Fwrites, knowledge-graph operations,\ncross-wing navigation, drawer management, and agent diaries. Installation\nand the full tool list:\n[mempalaceofficial.com\u002Freference\u002Fmcp-tools](https:\u002F\u002Fmempalaceofficial.com\u002Freference\u002Fmcp-tools.html).\n\n## Agents\n\nEach specialist agent gets its own wing and diary in the palace.\nDiscoverable at runtime via `mempalace_list_agents` — no bloat in your\nsystem prompt:\n[mempalaceofficial.com\u002Fconcepts\u002Fagents](https:\u002F\u002Fmempalaceofficial.com\u002Fconcepts\u002Fagents.html).\n\n## Auto-save hooks\n\nTwo Claude Code hooks save periodically and before context compression:\n[mempalaceofficial.com\u002Fguide\u002Fhooks](https:\u002F\u002Fmempalaceofficial.com\u002Fguide\u002Fhooks.html).\n\nFor per-message recall on top of the file-level chunks the hooks produce,\nrun `mempalace sweep \u003Ctranscript-dir>` periodically — it stores one\nverbatim drawer per user\u002Fassistant message, idempotent and resume-safe.\n\n---\n\n## Requirements\n\n- Python 3.9+\n- A vector-store backend (ChromaDB by default)\n- ~300 MB disk for the default embedding model\n\nNo API key is required for the core benchmark path.\n\n## Docs\n\n- Getting started → [mempalaceofficial.com\u002Fguide\u002Fgetting-started](https:\u002F\u002Fmempalaceofficial.com\u002Fguide\u002Fgetting-started.html)\n- CLI reference → [mempalaceofficial.com\u002Freference\u002Fcli](https:\u002F\u002Fmempalaceofficial.com\u002Freference\u002Fcli.html)\n- Python API → [mempalaceofficial.com\u002Freference\u002Fpython-api](https:\u002F\u002Fmempalaceofficial.com\u002Freference\u002Fpython-api.html)\n- Full benchmark methodology → [benchmarks\u002FBENCHMARKS.md](benchmarks\u002FBENCHMARKS.md)\n- Release notes → [CHANGELOG.md](CHANGELOG.md)\n- Corrections and public notices → [docs\u002FHISTORY.md](docs\u002FHISTORY.md)\n\n## Contributing\n\nPRs welcome. See [CONTRIBUTING.md](CONTRIBUTING.md).\n\n## License\n\nMIT — see [LICENSE](LICENSE).\n\n\u003C!-- Link Definitions -->\n[version-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fversion-3.3.3-4dc9f6?style=flat-square&labelColor=0a0e14\n[release-link]: https:\u002F\u002Fgithub.com\u002FMemPalace\u002Fmempalace\u002Freleases\n[python-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.9+-7dd8f8?style=flat-square&labelColor=0a0e14&logo=python&logoColor=7dd8f8\n[python-link]: https:\u002F\u002Fwww.python.org\u002F\n[license-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-MIT-b0e8ff?style=flat-square&labelColor=0a0e14\n[license-link]: https:\u002F\u002Fgithub.com\u002FMemPalace\u002Fmempalace\u002Fblob\u002Fmain\u002FLICENSE\n[discord-shield]: https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fdiscord-join-5865F2?style=flat-square&labelColor=0a0e14&logo=discord&logoColor=5865F2\n[discord-link]: https:\u002F\u002Fdiscord.com\u002Finvite\u002FycTQQCu6kn\n","MemPalace 是一个开源的AI记忆系统，用于存储和检索对话历史。其核心功能包括逐字存储文本、使用语义搜索进行检索，并且支持可插拔的后端架构，默认使用ChromaDB作为索引层。项目结构化地组织了数据，使得搜索可以更加精准高效。MemPalace 适用于需要本地优先处理AI记忆的应用场景，如个人或团队的知识管理、开发者的代码记忆等。用户可以在不依赖任何云服务或API的情况下，实现高精度（96.6% R@5）的记忆检索。",2,"2026-06-11 02:38:34","top_all"]