# Atomic
Atomic is a **local-first, AI-native [[Personal Knowledge Management (PKM)]]** system that turns markdown notes into a semantically-connected, AI-augmented knowledge graph. Notes are "atoms" that get automatically chunked, embedded, semantically tagged, and intelligently linked. Free and open-source (MIT).
## Key points
- **Atoms** — markdown notes auto-processed: chunking, embedding, LLM tagging, smart linking.
- **Semantic search** — vector embeddings over SQLite (sqlite-vec); find by meaning, not keywords. See [[Semantic Search]].
- **Wiki synthesis** — LLMs generate articles from grouped atoms with inline citations.
- **Spatial canvas** — force-directed graph laid out by semantic similarity.
- **Agentic chat** — query your notes without hallucinated answers (grounded, [[Retrieval-Augmented Generation (RAG)|RAG]]-style).
- **Capture** — browser extension and RSS feed → automatic atom creation.
- **[[Model Context Protocol (MCP)]] server** — exposes the knowledge base to Claude, Cursor, and other AI tools.
- **Clients** — desktop ([[Tauri]]), web (React), native iOS, CLI; self-host via Docker / Fly.io / Rust binary.
- **AI backends** — OpenRouter, Ollama, OpenAI-compatible APIs.
- **Stack** — [[Rust]] core (atomic-core), actix-web server, SQLite + sqlite-vec.
## Why it matters
Atomic is an AI-native take on the same space as [[Obsidian]] — but where Obsidian is files-first with AI bolted on, Atomic makes the embedding/graph/synthesis layer the product, while staying [[Local-First Software|local-first]] and self-hostable. Conceptually adjacent to [[Graphify]] (both build queryable knowledge graphs and expose them over MCP), but aimed at notes/PKM rather than codebases.
## References
- Website: https://atomicapp.ai/
- Repository: https://github.com/kenforthewin/atomic
## Related
- [[Personal Knowledge Management (PKM)]]
- [[Local-First Software]]
- [[Obsidian]]
- [[Model Context Protocol (MCP)]]
- [[Graphify]]
- [[Rust]]