Pi
A minimal terminal coding agent extended with skills, prompt templates and TypeScript packages
Install
npm install -g --ignore-scripts @earendil-works/pi-coding-agentAfter installing, run pi and use /login or set a provider key.
This is third-party code. Review the repository files before installing.
What it does
Pi gives the model four base tools: read, write, edit and bash. Everything else comes from TypeScript extensions, skills, prompt templates and themes, which can be bundled into packages and shared via npm or git. It ships without subagents or plan mode: you build them or install a package. The agent runs interactively, in print or JSON mode, over RPC and as an SDK. The repository also holds libraries for building agents: a unified multi-provider model API, an agent runtime and a terminal UI.
Who it is for. For developers who want a small agent and are willing to shape it around their workflow.
Good fit when
- You want a lean agent that is easy to extend
- You want to use a Claude, ChatGPT or Copilot subscription in one terminal agent
- You need to embed an agent in your app via RPC or SDK
Not a fit when
- You need built-in limits on file, network and credential access
- You need subagents and plan mode out of the box
Example request
Build yourself an extension that shows me each bash command and waits for confirmationLimitations
There is no built-in permission system: the agent runs with the launching user's rights, and the authors suggest a container or sandbox for isolation. Models connect via a provider subscription or API key, so VPN needs depend on the provider. Issues and pull requests from new contributors are auto-closed by default.
How to disable. Remove the global package with npm uninstall -g @earendil-works/pi-coding-agent.
Security check
- Runs arbitrary bash commands without built-in restrictions
- Has access to all of the user's files and credentials
README in short
The root README describes the monorepo: the coding agent, an agent runtime, a unified model provider API, a terminal UI and supporting packages. It states the agent has no permission system and lists three isolation options. It spends a lot of space on npm supply-chain hardening. The agent package README shows install via npm or a script and the provider list. MIT licensed.
FAQ
Which models are supported?
Claude Pro and Max, ChatGPT Plus and Pro, and GitHub Copilot subscriptions, plus API keys for Anthropic, OpenAI, Google, DeepSeek, Mistral, Groq and other providers.
How do I restrict the agent?
Run it in Docker or a sandbox; options are described in the containerization docs in the repository.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Anthropic's coding agent for the terminal, IDE and GitHub: understands your codebase, runs tasks and handles git