Pi

Pi is an open-source AI agent toolkit with 116,000+ stars on GitHub that bundles a unified LLM API, an agent loop, a terminal UI and a ready-to-use coding agent CLI.

Overview

Building an agent usually means stitching together a model client, a tool-calling loop, streaming handling and some kind of interface before you can test a single idea. Pi ships those pieces as one coherent toolkit. Its unified API smooths over the differences between LLM providers, so the same agent code can talk to different models without rewrites. On top of that sits a production-shaped agent loop that handles tool execution, message flow and state, a terminal user interface for interactive sessions, and a coding agent CLI you can use directly for software tasks. The project is MIT licensed and under very active development, with repository updates landing in October 2026. For developers who build agents rather than just use them, Pi is a batteries-included starting point that stays hackable all the way down.

Key Features

• Unified LLM API – one interface across providers; swap models without rewriting agent code.
• Agent loop included – tool calling, streaming and state management handled for you.
• Coding agent CLI – a working terminal coding agent out of the box, ready to extend.
• Terminal UI – interactive TUI for running and supervising agent sessions.
• Toolkit, not a black box – compose the pieces you need and inspect every layer.
• MIT licensed – permissive open source for commercial and personal projects alike.

💳 Pricing

Pi is completely free and open source under the MIT license. Your only cost is the LLM API usage of whichever provider you connect.

Who Is It For

Developers building custom AI agents and coding tools who want a unified model API and a solid agent loop without adopting a heavyweight framework.