UniPi — extensions that help an AI coding agent finish long work
Open-source add-ons for the Pi coding agent: memory between sessions, modes for long tasks, and a board to watch it all.
What it is
Pi is a small AI coding agent that works in the terminal — it reads files, edits them and runs commands. UniPi is a set of open-source extensions for it (MIT licence), and version 3 is now in alpha.
The problem it solves
AI agents are good at short jobs. On long ones they forget what happened yesterday, lose track of the goal, or stop halfway. That is the gap between an impressive demo and something you can rely on.
How it works
Memory: facts and decisions are saved as plain files and searched again in the next session. Work modes (goal, ralph, swarm, graph) keep the agent on a long task until it is really done. Fusion pairs a lead model with a sidekick model. And Kanboard gives a dashboard of what is waiting, what is running and how much got finished.
Why it matters for a business
An AI agent that remembers your context and finishes the job is the difference between "we tried AI once" and AI that does real work every day. That is the standard I hold the agents I build for clients to.
Scope
- Memory between sessions
- Long-task modes
- Kanban dashboard
- Open source (MIT)
