How to explore the pi-mono repository?

HEAD

3c9c54dAdd [Unreleased] section for next cycle

This post might have stale content, as HEAD is 338 commits ahead.

Julian Benegas posted

explore this repo

Claude Haiku 4.5 commented

Perfect! Now let me create a comprehensive summary of this repository:

##Repository Overview: pi-mono

This is the Pi Monorepo - a comprehensive toolkit for building AI agents with a focus on coding tasks. The project is maintained by Mario Zechner (@badlogic).

###Core Purpose

Pi is a framework for building AI agents with:

Unified multi-provider LLM API supporting OpenAI, Anthropic, Google, and many others
Agent runtime with tool calling and state management
Interactive coding agent CLI for real-world development tasks
Terminal UI (TUI) library with differential rendering
Web UI components for chat interfaces

###5 Main Packages

Package	Purpose
@mariozechner/pi-ai	Unified multi-provider LLM API supporting OpenAI, Anthropic, Google, Mistral, Bedrock, and 20+ more providers with advanced features like prompt caching, reasoning, extended thinking
@mariozechner/pi-agent-core	Core agent runtime with tool calling, state management, and extensibility for building custom agents
@mariozechner/pi-coding-agent	Production-ready interactive CLI agent for coding tasks with file editing, bash execution, git, debugging
@mariozechner/pi-tui	Terminal UI library with differential rendering, keybindings, autocomplete, and rich terminal features
@mariozechner/pi-web-ui	Web components for building AI chat interfaces

###Key Features

Multi-Provider Support: Abstracts 20+ LLM providers (OpenAI, Anthropic, Google, Mistral, Bedrock, Fireworks, OpenRouter, etc.)
Advanced LLM Features:
- Prompt caching (Anthropic, OpenAI)
- Reasoning/Extended thinking (Anthropic, OpenAI, Google)
- Tool calling standardization
- Image and file handling
Coding Agent Capabilities:
- File editing with context
- Bash command execution
- Git integration
- Interactive and batch modes
- Session management with branching and compaction
Session Recording & Sharing: Can publish sessions to Hugging Face for improving agents with real-world data
Extensible Architecture: Custom providers, extensions, and skills for the coding agent

###Development Stack

Language: TypeScript
Runtime: Node.js (v20+)
Build Tools: Biome (linting/formatting), tsx, vitest
Monorepo: npm workspaces

###Project Stats

~300 files across 5 packages
Extensive test suite including unit, integration, and E2E tests
80+ test files covering providers, agents, UI, and tools
Active development with CI/CD via GitHub Actions

###Notable Commands

npm install          # Install all dependencies
npm run build        # Build all packages
npm run check        # Lint, format, and type check
./test.sh            # Run tests
./pi-test.sh         # Run pi from source (from any directory)
npm run dev          # Watch mode for development

###Contribution Model

The project has a quality-first gatekeeping system:

New issues/PRs auto-closed; maintainers review daily
lgtmi approval: future issues not auto-closed
lgtm approval: issues + PR rights
Must follow CONTRIBUTING.md and AGENTS.md
Core rule: You must understand your code; no un-reviewed AI-generated slop
Uses agents for development but enforces review practices

###Community

Discord: https://discord.com/invite/3cU7Bz4UPx
Hugging Face sessions: https://huggingface.co/datasets/badlogicgames/pi-mono
Chat bot workflows: https://github.com/earendil-works/pi-chat

What would you like to explore further? I can dive deeper into any specific package, architecture patterns, the agent implementation, provider support, or anything else in the codebase.

END OF POST