wtf is pi-agents?

Pi Agents is an independent version of Pi, a coding agent tool, built around a persistent workspace for managing agents. When you start it plainly, it opens straight into this workspace, and the individual worker sessions it manages still behave like normal Pi sessions underneath, keeping their full conversation history intact. The tool keeps its own internal state in a hidden folder under your home directory, separate from the regular Pi configuration. The project is organized into several packages. One provides a single way to talk to different AI providers such as OpenAI, Anthropic, and Google through one interface. Another is the core agent runtime that handles calling tools and keeping track of state. A third is the interactive command line coding agent itself, which is where the persistent Agents workspace lives. There is also an experimental package for orchestrating multiple agents together, still under active development, and a terminal interface library used to render the command line screens efficiently. The README is upfront that Pi Agents does not include any built in system for limiting what the agent can access. By default it runs with whatever permissions the user who started it already has, meaning it can read and write files, run processes, and reach the network without extra restriction unless you set that up yourself outside the tool. Beyond this, the README is fairly brief. It gives basic instructions for cloning the repository, installing dependencies without running their setup scripts, building the packages, and running checks and tests, but it does not go into detail about day to day usage beyond pointing to a separate documentation file inside the coding agent package. The project is written in TypeScript and released under the MIT license.