Three lines to a memory
every tool reads
SecondOS is a Bun monorepo with a CLI, an MCP server and a GitHub Action for freshness. Install, point it at a repo, connect a tool. That is the whole loop.
The three-line start
# 1 — install + sync (builds the map on your machine) bunx @secondos/cli push . # 2 — point it at your repo (that's it — re-runs stay incremental) # the map + embeddings upload, never your source # 3 — connect an AI tool over MCP (Cursor, Claude Code, Copilot…) bunx @secondos/cli mcp-config # prints the config to paste into your editor
Read next
Quickstart
Install, sync your repo, and connect your first AI tool over MCP. The first sync is a minute or two on a small repo, closer to ten on a big one. After that it's incremental. No credit card.
Open →MCP server
The tools your AI tools get: codebase (search, overview, who_uses, impact_of, drift), memory, and context.
Open →Remote MCP
Connect a hosted agent (claude.ai, ChatGPT, any MCP-over-HTTP client) with a project-scoped, least-privilege token.
Open →Living knowledge base
Eleven docs auto-written from your map and memories and kept current on every push. Read them in the web app or over MCP via read_living_doc.
Open →CLI
The commands: push, ask, drift, mcp-config, init-github-action. Deep sync, offline mode, and how re-sync stays incremental.
Open →How we measure
Our real numbers and the methodology behind them: memory recall, code search, and relationship-graph resolution on real repos, a precision stance, never a guess.
Open →Ready? Point it at a repo.
> daemon: three lines. connect your repo, I'll handle the rest.