pi-mono
Properties
created
26.04.2026, 20:37
modified
26.07.2026, 13:50
published
Empty
sources
badlogic/pi-mono
topics
Agent Harness, Coding Agent, LLM API
authors
claude-sonnet-4-6
ai-assisted
Yes
- badlogic/pi-mono
- 31.1k stars
- Monorepo by Mario Zechner containing an opinionated, minimal AI agent toolkit
- Philosophy: build only what’s needed — no MCP, no built-in todos, no plan mode, no sub-agents, no background bash
- Runs on Windows, Linux, macOS (Node.js runtime)
# Videos
# Packages
- pi-ai
- Unified LLM API with multi-provider support (Anthropic, OpenAI, Google, xAI, Groq, Cerebras, OpenRouter, any OpenAI-compatible endpoint)
- Streaming, tool calling with TypeBox schemas, thinking/reasoning support
- Cross-provider context handoffs, token and cost tracking
- Works in the browser (Anthropic and xAI support CORS)
- pi-agent-core
- Agent loop: tool execution, validation, event streaming
Agentclass: state management, message queuing, attachment handling, transport abstraction
- pi-tui
- Minimal terminal UI framework with differential rendering
- Synchronized output (
CSI ?2026h/l) for flicker-free updates - Components: editors with autocomplete, markdown rendering
- Works well in Ghostty / iTerm2; some flicker in VS Code terminal
- pi-coding-agent
- CLI coding agent wiring all the above together
- Session management (continue, resume, branching), custom slash commands, OAuth for Claude Pro/Max
- Minimal system prompt (~1000 tokens total), minimal toolset (read, write, edit, bash)
- YOLO mode by default — no permission prompts, no safety rails
- HTML session export, headless JSON streaming / RPC mode
# Related
- What I learned building an opinionated and minimal coding agent
- Author’s blog post detailing design decisions, lessons learned, and benchmark results
- Building Pi With Pi
- 2026-05-24 — Armin Ronacher on dogfooding Pi: how clanker-generated issues and over-engineered fixes change the maintainer’s role