earendil-works/pi

What it solves

Pi provides a modular framework for building and running AI agents, specifically focusing on a self-extensible coding agent that can interactively solve programming tasks via a CLI.

How it works

The project is structured as a monorepo of specialized packages:

• pi-ai: A unified API that allows the agent to communicate with multiple LLM providers (such as OpenAI, Anthropic, and Google).
• pi-agent-core: The runtime engine that handles state management and tool calling.
• pi-coding-agent: An interactive CLI implementation of a coding agent.
• pi-tui: A terminal user interface library for rendering output.
• pi-telemetry: A system for vendor-neutral telemetry and typed schemas.

To ensure security, the project supports various containerization patterns (like Docker or micro-VMs) to sandbox the agent's filesystem and network access.

Who it’s for

Developers who want a powerful, extensible coding agent or a modular runtime for creating their own AI agents with multi-provider LLM support.

Highlights

• Multi-provider LLM support: Unified interface for OpenAI, Anthropic, and Google.
• Self-extensible: Designed to be extended by the agent itself.
• Modular architecture: Separated core runtime, AI API, and CLI interface.
• Sandboxing options: Built-in guidance for running agents in Docker or micro-VMs for security.