earendil-works/pi

Project overview

A self-hosted coding agent and CLI framework that unifies multiple hosted LLM APIs (OpenAI, Anthropic, Google) into a single interface. Requires external API credentials and lacks a built-in permission system, making external sandboxing necessary for restricted environments.

Project type

AI Agent · AI Coding · Infrastructure

Use cases

Coding & Development · Automation

Deployment

Refer to project documentation

License

MIT

Best for

• Developers who need a locally run, self-hosted framework to build and run coding agents via a terminal interface using external hosted LLMs.

Key capabilities

• An interactive coding agent that operates directly within the terminal environment to execute tasks.
• A single API to communicate with large language models from various providers, including OpenAI, Anthropic, and Google.

Limitations and risks

• The project does not include a built-in system to restrict filesystem, process, network, or credential access.
• By default, the agent runs with the permissions of the user and process that launched it, presenting a risk if not externally contained.

Getting started

• Installation requires Node.js and npm package installation or running from source. Users must configure external LLM provider API credentials before running the CLI to achieve first success.

Evidence and sources

• GitHub project description: AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI
• README: This is the home of the Pi agent harness project including our self extensible coding agent.
• README: | **[@earendil-works/pi-ai](packages/ai)** | Unified multi-provider LLM API (OpenAI, Anthropic, Google, etc.) |
• README: Pi does not include a built-in permission system for restricting filesystem, process, network, or credential access. By default, it runs with the permissions of the user and proce…
• README: - **Plain Docker**: run the whole `pi` process in a local container for simple isolation.

Reviewed at: 2026-07-17T05:19:26Z