Atomic

Atomic is a self-hosted personal knowledge base (MIT-licensed) that turns plain Markdown notes into a semantically connected, AI-augmented knowledge graph. The desktop client is built with Tauri (Rust + TypeScript) and saves notes as "atoms" — markdown files that are automatically chunked, embedded, and tagged. Where Notion treats pages as a hierarchy, Atomic treats them as a graph you explore on a force-directed canvas and query through semantic vector search.

When you save a note, Atomic chunks the markdown, generates vector embeddings, and uses an LLM to auto-tag and link the atom to existing notes by semantic similarity. The result is a sqlite-vec index that powers search-by-meaning rather than keyword matching, and a canvas view that arranges related atoms spatially based on their embeddings. You can also trigger Wiki Synthesis to have the LLM generate cited articles composed from across your atom cluster.

Atomic is provider-agnostic — embeddings and LLM calls go through OpenRouter, Ollama, or any OpenAI-compatible endpoint, so you can use a local Ollama model for full privacy or pay-as-you-go via OpenRouter. The built-in MCP server even exposes your knowledge base to external agents like Claude. An optional hosted version, Atomic Cloud, is offered for users who don't want to run Docker themselves.

Atomic ships as a Tauri desktop app for macOS, Linux, and Windows and as a headless server with a separate web frontend. The recommended self-host path is a single docker compose up -d that brings up the API, the web client, and an nginx reverse proxy listening on :8080. A fly.toml.example is included for one-command deployment to Fly.io with persistent volumes, and a native iOS app is on the App Store while a Chrome extension captures web content directly into your knowledge base.