Skip to content
aicoolies logo
pi logo

Pi

Minimal terminal coding harness

Pi Coding Agent is an MIT-licensed Node.js CLI from earendil-works for building and running coding agents in a local terminal. The current package describes a read/bash/edit/write toolset and session management, while the repo positions Pi as a unified LLM API, agent loop, TUI, and coding-agent CLI. It is best framed as a lean, self-extensible BYO-model toolkit rather than a managed IDE.

About Pi

Pi Coding Agent is an MIT-licensed terminal coding-agent toolkit now canonical at github.com/earendil-works/pi. GitHub redirects the older badlogic/pi-mono path to the earendil-works/pi repository, which currently shows 66K+ stars, active pushes, and release v0.80.2. The npm package @earendil-works/pi-coding-agent is also at 0.80.2 and describes a coding-agent CLI with read, bash, edit, write tools and session management.

The safest product framing is not a polished managed IDE or enterprise admin layer. Pi is a local Node.js CLI and agent toolkit for developers who want an inspectable loop, terminal UI, and bring-your-own-model workflow they can reason about and extend. The repository description positions it as a unified LLM API, agent loop, TUI, and coding-agent CLI, while pi.dev and package metadata reinforce the terminal-agent orientation rather than a hosted SaaS promise.

That makes Pi most relevant for experienced developers and teams comparing lightweight coding-agent harnesses with larger tools such as Claude Code, OpenCode, or Cursor. Use it when local control, source visibility, and extensibility matter more than turnkey governance dashboards. Avoid brittle copy about exact prompt size, unsupported provider lists, or old repository paths; keep the page anchored to the canonical repo URL, MIT license, current npm/release version, and the read/bash/edit/write plus session-management toolset that public package metadata supports.

Pricing & Platform Specs

Pricing Summary

Free and 100% open source under the MIT license. Pi charges no subscription or license fees; developers bring their own API keys (BYOK) for cloud LLM providers or run local open-source models via Ollama and vLLM at zero compute license cost.

full pricing breakdown →

Supported Platforms

CLI (Node.js)

Explore categories, tags & use cases

Alternatives

All Pi alternatives →

Open-source AI coding agent for the terminal

Open-source terminal-based AI coding agent built in Go by the SST team, with a rich TUI (Bubble Tea) supporting 75+ model providers including OpenAI, Anthropic, Gemini, Bedrock, Groq, and OpenRouter. Features vim-like editing, persistent SQLite sessions, and LSP integration for 40+ languages. Fully free with no vendor lock-in, it has rapidly grown to 95k+ GitHub stars.

Open Source

Glamorous agentic coding for the terminal by Charm

LSP-enhanced terminal coding agent from Charmbracelet with the broadest cross-platform support including Android. Features mid-session model switching and MCP extensibility for custom tool integration. Combines the elegance of Charmbracelet's TUI toolkit with practical coding assistance, offering a polished terminal experience for developers on any platform.

Multi-agent terminal coding assistant

Codebuff is an open-source AI coding assistant that operates from the terminal, editing your codebase through natural language instructions while understanding project structure, dependencies, and patterns. Uses multi-agent architecture with specialized File Explorer, Planner, and execution agents for precise code modifications across files. Editor-agnostic via terminal-level operation. Supports any OpenRouter model including Claude, GPT, DeepSeek, and Qwen.

paid

Enterprise-grade AI coding agent system by Factory

System of specialized AI Droids — Code, Knowledge, Reliability, and Product — each optimized for specific development tasks. Ranked #1 on Terminal-Bench with 58.75% score. BYOK model with support for Anthropic and OpenAI models. Enterprise-focused approach that treats AI coding as a team of specialized agents rather than a single general-purpose assistant.

freemium

Side-by-Side Comparisons

pi logo
Pi
vs
Claude Code logo
Claude Code

Pi Coding Agent vs Claude Code: Minimal Agent Harness or Production Coding CLI?

Pi Coding Agent is a compact, MIT-licensed agent harness for developers who want to inspect and extend the coding-agent loop, while Claude Code is Anthropic's integrated coding-agent CLI with a stronger official product surface for professional teams. Claude Code serves as the more dependable production standard across software teams because it offers the more complete, documented, vendor-backed coding workflow; Pi is best for local experimentation, custom extensions, and agent-loop research.

Community experience

Sources & verification

Sources checked
Content verified

Verification dates are editorial checks. Routine CMS saves and automatic updatedAt timestamps do not advance them.

FAQ

What is Pi?

Pi Coding Agent is an MIT-licensed Node.js CLI from earendil-works for building and running coding agents in a local terminal. The current package describes a read/bash/edit/write toolset and session management, while the repo positions Pi as a unified LLM API, agent loop, TUI, and coding-agent CLI. It is best framed as a lean, self-extensible BYO-model toolkit rather than a managed IDE.

Is Pi free?

Yes — Pi is open source and free to use. Free and 100% open source under the MIT license. Pi charges no subscription or license fees; developers bring their own API keys (BYOK) for cloud LLM providers or run local open-source models via Ollama and vLLM at zero compute license cost.

Is Pi open source?

Yes — Pi is open source.

Is Pi still maintained?

Yes — Pi is active. Its listing was last verified on September 6, 2026.

What are the best Pi alternatives?

The first editor-selected Pi alternatives are OpenCode, Crush, Codebuff, and more.

How does Pi score in our review?

The published editorial review lists Pi at 84/100 overall across speed, privacy, and developer experience. Check the review's evidence status and test metadata for its verification level.