PoorMad documentation
Everything you need to install, run, extend, and self-host the self-improving AI agent.
PoorMad is a self-improving AI agent. The same agent core runs across a command line, a native desktop app, a web portal, and 20+ messaging platforms. It learns from experience by writing reusable skills, recalls past context through memory, and coordinates tools and subagents to get work done.
What PoorMad is
At its heart, PoorMad is one agent brain with many front doors. You can talk to it from a terminal, a desktop window, a browser, or Telegram — and it is the same agent with the same memory and skills behind all of them. That is possible because every surface connects to a single gateway using one JSON-RPC protocol.
Design principles
- Local-first. Memory and skills live in your PoorMad home directory. Nothing leaves your control unless you choose to connect a remote surface.
- Open source. MIT licensed, built in the open. Capability compounds locally as the agent writes its own skills.
- Mono-hue identity. One violet accent, no noise. Calm, fast, and yours.
- Learn by doing. The agent turns repeated work into skills it can reuse, so it gets cheaper and more capable over time.
Who it is for
Individuals and small teams who want a personal agent they actually own — developers wiring it into their stack, operators running it on a server, and anyone who would rather self-host than hand their context to a third party.
Next steps
Install it in minutes with Installation, run your first task in Quickstart, or understand the moving parts in Architecture.