agent-native-agent
Self-hosted apps you operate by watching a dashboard and talking to a coding agent that IS the runtime — it proposes, applies, and evolves the app live. Built f
What is agent-native-agent?
Self-hosted apps you operate by watching a dashboard and talking to a coding agent that IS the runtime — it proposes, applies, and evolves the app live. Built for Claude Code and MCP.
Overview of agent-native-agent
What it is
ANA (Agent-Native Agent) is a self-hosted application that you operate by watching a live dashboard and conversing with a coding agent that serves as the runtime. It is built for Claude Code and MCP. The agent proposes changes, applies them on approval, and evolves the app at runtime. There are two editions: Agent-Native Agent (this repository, licensed under AGPL-3.0) and Agent-Native Agent Pro (commercial). The base edition includes six tabs: PRD, Work, Knowledge, Requests, Evolve, and Update. The Pro edition adds a knowledge layer, graph view, system scope with admin and member roles, roadmap and health, multiple workspaces, a Claude account per session, and an offline edition.
Key features
- Operate by watching a dashboard and talking to the agent – no clicking through fixed UI.
- Agent as runtime: the agent reads your data, acts, and rewrites the app's own code when asked.
- Self-hosted with zero dependencies; runtime state lives in
.ana/(git-ignored). - No bridge or MCP: browser posts to server, server injects text into a tmux pane, and a 300 ms capture-pane loop mirrors the session as SSE.
- Proposals: the agent posts a before/after diff; on approval the server applies the diff and bumps a version so all devices re-sync.
- Naming convention: folder name becomes the ANA name; tmux sessions are named
-ana-. - Base starts empty – you tell ANA what your work is and it writes the PRD.
Use cases
- Building self-hosted apps that you operate by watching and talking.
- Collapsing the gap between using and building – the same conversation changes the app.
- Anyone who wants to own their harness and have the agent as the runtime, evolving the app live.
- Suitable for teams or individuals who need a malleable, self-hosted tool that grows with their work.
Getting started
Prerequisites: Node ≥ 24, tmux, and a coding-agent CLI (e.g., Claude Code). Manual run: bash git clone https://github.com/tykimos/agent-native-agent ~/ana/base-ana cd ~/ana/base-ana # 1) start your coding agent inside tmux tmux new -s base-ana-claude # inside the session, run: claude # 2) in another terminal, start ANA node server.js # → http://localhost:8809 Open http://localhost:8809 and tap the chat button. The server configures the tmux pane when it connects. Runtime state lives in .ana/. On first run the board is seeded with two small example flows (ANA_SEED=0 to start empty). For automated setup, the install skill scripts (check-env.sh, install.sh, run.sh) are provided in the repository.
License and pricing
The Agent-Native Agent repository is licensed under AGPL-3.0 (open source). Agent-Native Agent Pro is commercial; details are available in COMMERCIAL.md in the repository.
Based on the project's public README and repository metadata.