All your AI conversations, one graph, one chat.
Status: pre-alpha. Importing and search work today: ChatGPT, Claude, and Gemini exports, Claude Code and Codex sessions, notes, documents (PDF, Word, Excel, and more), email, and archives of any of them land in a local library and a SQLite graph store you can search from the terminal by words, by meaning, or both. A language model turns them into a knowledge graph of entities, relationships, and topics that you can browse and search, with facts that each lead back to the message that said them, and you can ask questions and get answers with sources, in the terminal, over a REST API, or in a web interface where the knowledge graph is a way into your chat history. AI assistants such as Claude and Cursor can search it too, through an MCP server. A library moves between machines as one archive file. The roadmap below shows what comes next.
With uv installed, one command downloads ChatLore and opens it on a made-up library:
uvx chatlore demoThe demo holds 32 invented conversations with their knowledge graph already
built, so search, topics, and the graph explorer work at once, with no export
and no API key. It lives in ~/.chatlore-demo, apart from your own library.
Asking questions also needs a model key; see docs/models.md.
uv tool install chatlore # or: pipx install chatlore
chatlore import path/to/chatgpt-export.zip ~/Documents/notes # exports, documents, archives
chatlore search "postgres index"
chatlore process # chunk and embed, local model
chatlore search "why was my query slow" --semantic
chatlore search "slow postgres query" --hybrid # words and meaning together
chatlore extract --limit 50 # entities, needs a model key
chatlore topics # what the conversations are about
chatlore entity "postgres" # one entity and where it came up
chatlore facts "backups" # decisions and findings, with sources
chatlore ask "why was my query slow?" # an answer with sources
chatlore serve # web UI and API on http://127.0.0.1:8000
chatlore mcp # tools for Claude, Cursor, and other MCP clients
chatlore export chatlore.zip # the whole library in one file
chatlore statsThe library is kept in ~/.chatlore; --home <folder> or CHATLORE_HOME picks another.
The web interface can do the same without the terminal: after chatlore serve,
Your data uploads an export, builds everything from it, and downloads the
library again.
The source is detected from the file. Importing is idempotent, so re-running it after a fresh export only adds what changed. How to get each export, what is kept, and the known limits are in docs/importers.md. Extraction and chat use any OpenAI-compatible model, OpenRouter by default; see docs/models.md. The knowledge graph is described in docs/extraction.md, chat and the API in docs/chat.md, the web interface in docs/web.md, setting up assistants over MCP in docs/mcp.md, archives and Markdown export in docs/export.md, hosting the demo in docs/hosting.md, and keeping the graph in FalkorDB instead of SQLite in docs/falkordb.md.
ChatLore is a local-first, open-source graph knowledge base built from your own conversations and documents.
- Import your history from ChatGPT, Claude, and Gemini exports, Claude Code and Codex CLI sessions, Markdown folders, notes, and documents.
- Link everything into one graph of conversations, entities, topics, and facts, with every extracted fact pointing back to the message it came from.
- Search across all of it with full-text, vector, and graph retrieval combined.
- Chat in one place with citations, using any model provider you like: OpenAI-compatible endpoints (including Ollama, vLLM, and Qwen), Anthropic, or Gemini.
- Integrate with the tools you already use through an MCP server, a REST API, and a CLI.
Your data stays in a folder you own. Import and search work without any LLM; extraction is an optional enrichment you can re-run with a better model later.
| Milestone | Deliverable | Status |
|---|---|---|
| M0 | Repository skeleton and CI | done |
| M1 | Core data model and embedded SQLite graph store | done |
| M2 | Importers: ChatGPT, Claude, Gemini, Markdown, notes | done |
| M3 | Chunking, embeddings, hybrid search | done |
| M4 | Entity, topic, and fact extraction with provenance | done |
| M5 | REST API with streaming chat | done |
| M6 | Web UI: conversations, graph explorer, chat | done: Explore your chats through the graph, light and dark themes |
| M7 | MCP server for Claude Desktop, Claude Code, Cursor, ChatGPT | done; ChatGPT through /mcp once the demo is online |
| M8 | Easy to try: PyPI package, demo library, export and import | done, v0.1.0 |
| M9 | Hosted demo | public mode, MCP over HTTP, Docker image, and Hugging Face Space deployment done; going live next |
| M10 | FalkorDB backend | done |
| M11 | Your own data in the web interface: upload, export, private visitor libraries | done |
| M12 | Import anything: documents, email, data files, folders, and archives | done |
Requirements: uv and Git. uv installs the pinned Python version for you.
git clone https://github.com/cl0ver012/chatlore.git
cd chatlore
uv sync
uv run chatlore --helpChecks that CI runs on every pull request:
uv run ruff check .
uv run ruff format --check .
uv run mypy
uv run pytestSee CONTRIBUTING.md for branch naming, commit conventions, and the pull request checklist.