[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-79879":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":15,"subscribersCount":15,"size":15,"stars1d":15,"stars7d":15,"stars30d":16,"stars90d":15,"forks30d":15,"starsTrendScore":15,"compositeScore":17,"rankGlobal":10,"rankLanguage":10,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":21,"hasPages":19,"topics":22,"createdAt":10,"pushedAt":10,"updatedAt":33,"readmeContent":34,"aiSummary":35,"trendingCount":15,"starSnapshotCount":15,"syncStatus":36,"lastSyncTime":37,"discoverSource":38},79879,"dejavu","JSingletonAI\u002Fdejavu","JSingletonAI","Memory that follows you across every AI tool. No cloud storage. No account required. Set it up once, use it everywhere.","https:\u002F\u002Fdeja-vu.dev",null,"Python",34,3,1,0,32,1.81,"Apache License 2.0",false,"main",true,[23,24,25,26,27,28,29,30,31,32],"ai-agents","claude","llm","local-first","mcp","memory","privacy","rag","sqlite","venice","2026-06-12 02:03:55","# Deja Vu\n\n[![Website](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fwebsite-deja--vu.dev-0f172a?style=flat-square)](https:\u002F\u002Fdeja-vu.dev)\n[![X](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F@useDejaVu-0f172a?style=flat-square&logo=x&logoColor=white)](https:\u002F\u002Fx.com\u002FuseDejaVu)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Flicense-Apache--2.0-0f172a?style=flat-square)](LICENSE)\n\u003Cimg width=\"1983\" height=\"793\" alt=\"GithubImage2\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F831be4fe-7384-428b-8e97-864c22255d84\" \u002F>\n\nLocal-first AI memory for agents and assistants.\n\nYour AI tools forget everything between sessions. The ones that don't store your context on someone else's servers. Deja Vu is the third option — a memory layer that runs on your machine, in SQLite, and plugs into anything that speaks Python, REST, CLI, or MCP.\n\n**One memory store, every tool.** Add a preference from the CLI, retrieve it in Claude Desktop, query it from a Python agent — same database, no sync, no account. Context built up in one tool is immediately available in the next.\n\n**Private by default.** Memories live in `~\u002F.dejavu` on your machine. The only thing that leaves is the LLM call itself, routed through Venice's privacy-focused API. No telemetry, no hosted memory service, no vendor lock-in. Open the SQLite file with any client and inspect every byte.\n\n## Quick start\n\n```bash\npip install dejavu-memory\ndejavu init\ndejavu add \"I prefer concise technical explanations\"\ndejavu search \"How should responses be written for me?\"\n```\n\nThat's it. Memories are saved to `~\u002F.dejavu\u002F` and searchable immediately.\n\n### Set your Venice API key\n\nDeja Vu uses Venice for memory extraction and reasoning. Get a key at [venice.ai](https:\u002F\u002Fvenice.ai) and export it:\n\n```bash\nexport VENICE_API_KEY=\"your-key\"\n```\n\n## Python\n\n```python\nfrom dejavu import Memory\n\nmemory = Memory()\nmemory.add(\"I prefer local-first tools.\", user_id=\"local_user\")\n\nresults = memory.search(\"What tools do I prefer?\", user_id=\"local_user\")\nprint(results)\n```\n\n## Local REST API\n\n```bash\ndejavu serve\ncurl http:\u002F\u002F127.0.0.1:8765\u002Fhealth\n```\n\nRuns on `127.0.0.1` by default — not exposed to your network.\n\n## How it works\n\nDeja Vu is three layers: an interface layer (Python, CLI, REST, MCP), a memory engine that extracts and ranks what's worth remembering, and a local SQLite store. Venice handles the LLM calls for extraction and search — everything else runs on your machine.\n\u003Cimg width=\"929\" height=\"695\" alt=\"image\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002Feb7aed95-eac3-48a8-8fde-fdecfdd8b1e8\" \u002F>\n\nWhen you add a memory, the engine asks Venice to pull out durable facts and preferences from the raw text, then writes them to SQLite with embeddings. When you search, it embeds the query, pulls the closest matches, and optionally re-ranks them through Venice for relevance.\n\nEvery interface hits the same engine and the same store. Add a memory through the CLI, retrieve it from Claude Desktop over MCP — same database, no sync.\n\n## MCP\n\nRun Deja Vu as an MCP server so local agents and editors share the same memory store.\n\n```json\n{\n  \"mcpServers\": {\n    \"dejavu\": {\n      \"command\": \"dejavu\",\n      \"args\": [\"mcp\"],\n      \"env\": {\n        \"VENICE_API_KEY\": \"your-key\"\n      }\n    }\n  }\n}\n```\n\nDrop this into your Claude Desktop, Cursor, or any MCP-compatible client config.\n\n## Interfaces\n\n| Interface | Use it for |\n| --- | --- |\n| Python SDK | Embedding memory directly into agents and scripts |\n| CLI (`dejavu`) | Quick adds, searches, and inspection from the terminal |\n| REST API | Language-agnostic access, local services, internal tools |\n| MCP server | Sharing one memory store across Claude Desktop, Cursor, and other MCP clients |\n\n## Privacy\n\n- Memories stored locally under `~\u002F.dejavu`\n- No hosted memory account required\n- Telemetry off by default\n- LLM calls go through Venice only\n\n## Project structure\n\nDeja Vu writes everything to `~\u002F.dejavu\u002F`:\n\n```\n~\u002F.dejavu\u002F\n├── config.json   ← Venice key, model choices, local settings\n├── memories.db   ← SQLite store: memories, embeddings, metadata\n└── logs\u002F         ← request logs (off by default, opt-in via config)\n```\n\nThe repo itself:\n\n\u003Cimg width=\"538\" height=\"798\" alt=\"image\" src=\"https:\u002F\u002Fgithub.com\u002Fuser-attachments\u002Fassets\u002F1658bf57-20a0-4560-bf48-dc4338666b25\" \u002F>\n\n```\ndejavu\u002F           ← core Python SDK and local memory engine\ncli\u002F              ← Python and Node CLIs\ndocs\u002F             ← documentation\nexamples\u002F         ← demo apps and integration samples\ntests\u002F            ← SDK and interface tests\n```\n\n## Background\n\nThe goal is to make AI memory feel boring and dependable: easy to run, easy to inspect, and easy to move between tools. The design follows the white paper's long-term memory direction while keeping the first public version focused on the parts people can actually use right away.\n\n## Attribution\n\nDeja Vu is based on the open-source work from [`mem0ai\u002Fmem0`](https:\u002F\u002Fgithub.com\u002Fmem0ai\u002Fmem0) and is licensed under Apache-2.0.\n","Deja Vu 是一个本地优先的AI记忆层，旨在让用户的记忆在不同的AI工具间无缝跟随。项目使用SQLite作为本地存储，并通过Python、REST、CLI或MCP接口与各种AI代理和助手集成。其核心功能包括无需云存储和账户即可实现的记忆管理，以及通过Venice API进行记忆提取和搜索的能力。特别适用于重视隐私且希望避免数据被第三方存储的用户场景中，如个人开发者、研究者或是对数据安全有较高要求的企业环境。此外，由于所有操作都在本地完成，因此能够有效防止供应商锁定。",2,"2026-06-11 03:58:22","CREATED_QUERY"]