Search

Search IconIcon to open search

pi-mono

Last updatedUpdated: by Jakub Žovák · 2 min read

Properties
created 26.04.2026, 20:37
modified 26.07.2026, 13:50
published Empty
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
    • Agent class: 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