Open-source AI work agent (MIT, self-hosted). Coding-agent-grade loop for long-horizon work — code, docs, slides, sheets, multi-agent, MCP & WeChat/Feishu on one Turn Log.
README
Open-source AI Work Agent — coding-agent-grade loop for long-horizon work. Self-hosted, multi-agent, MIT.
A human–AI co-thinking workspace: the same Agent Loop handles files, shell, and multi-agent coding, and co-edits docs, slides, and sheets on Document Stage (propose → review → keep/revert). Every Tool Call is logged — resume, replay, or edit the trail and continue.
Code, reports, decks, sheets, demos, automations — from Web, Desktop, CLI/TUI, or WeChat / Feishu / WeCom / QQ — on one trail.
| | |
|--|--|
| Positioning | General-purpose Work Agent — coding is a lane, not the ceiling |
| Control | Pure LLM-driven — no hand-maintained graph / role router / mode |
| Abstraction | Everything is a Tool — delegate_agent, ask_user, memory, table store, MCP… |
| State | Log is state — Turn Log → recover, replay, edit a result and continue |
| Office | Document Stage + AI Diff (Keep / Revert / hunks); text stays source of truth |
| Surfaces | Web · Desktop · CLI · TUI · IM channels · Document Stage |
MIT · Anthropic & OpenAI-compatible providers · Local-first under ~/.danmo-work/
| Platform | Package |
|----------|---------|
| macOS (Apple Silicon) | Homebrew (recommended) or .dmg |
| Windows | Setup .exe |
| Linux (x86_64) | AppImage / .deb |
All binaries: GitHub Releases.
brew tap danmo-ai/tap
brew in