What is pi-fleet?

pi-fleet is a command line tool for controlling many AI coding agents at once, specifically ones built on a system called Pi. Instead of watching each agent run in its own terminal window, which becomes impractical once you have dozens or hundreds running, pi-fleet gives you a way to create, message, check on, and shut down these agents through simple commands that return structured data your own programs can read. The project positions itself as different from terminal tools like tmux, which just show you the raw output of a running process. pi-fleet instead understands the specific way Pi agents work: whether an agent is currently busy or waiting, whether a message it was sent was actually accepted, and what its exact last response was. Other software, such as an orchestrator that assigns tasks to many agents or a dashboard that tracks their progress, is expected to be built on top of pi-fleet rather than included in it. Each agent you create gets a stable name you refer to it by, plus its own underlying Pi session, which is a record of everything said in that conversation. Importantly, pi-fleet does not own or modify these session files. They stay in Pi's normal storage format, so other tools can read or analyze them independently, and deleting an agent from pi-fleet never deletes its underlying session. Typical commands let you create a named agent, send it a task, wait for it to finish and receive its answer, check its current status, or watch its conversation update live. Every command that completes returns one clean block of structured data, making it straightforward to hook into scripts or other software. The tool currently only runs on Linux computers with a specific range of Node.js versions installed, and is installed through npm as a beta package. The full README is longer than what was shown.