Pi
Extensible open-source coding agent for the terminal from Austria that works with models from many providers or locally.
What it does
Pi is a coding agent for the terminal: it reads and edits files in your project, runs commands and works through tasks using a language model of your choice. It connects to many providers such as Anthropic, OpenAI, Google, Mistral, OpenRouter or Ollama, via API key or login. The core is deliberately small; features like sub-agents, a plan mode or confirmation prompts before actions are added through TypeScript extensions, skills and packages. Besides the interactive interface there is a JSON mode for scripts, an RPC interface and an SDK for embedding it in your own applications. MCP is supported.
Who it suits
Development teams that want to tailor their agent themselves and avoid being tied to a single model provider. If you are looking for a ready-made solution with a graphical interface, other tools are a better fit.
Data protection for businesses
Pi runs locally on your machine; code and prompts go to the model provider you configure. Server location, data processing agreement and whether prompts are used for training depend on that provider’s terms. With a local model, data stays within your own network. The vendor does not mention a privacy policy or a data processing agreement for Pi itself. There are no business plans; the Radius gateway with organisation features is still in an early alpha phase.
Limits
Pi has no built-in sandbox and does not ask for confirmation before tool calls. The agent runs with the permissions of your user account; the vendor recommends containers or virtual machines. Getting started is hard without terminal experience.