Gptme Open source

What it does

gptme is an open-source, terminal-based personal AI agent. It gives an LLM direct access to your machine — a shell, a Python REPL, and file read/write/patch tools — so it can carry out multi-step coding and system tasks interactively, in a fully autonomous "agent" mode, or inside CI/CD pipelines.

What is inside

Out of the box gptme wires up: shell execution, a Python interpreter, incremental file patching, web browsing/scraping via Playwright, vision (screenshots and image analysis), and a plugin system for adding custom tools, "skills," "lessons" and hooks. Persistent autonomous agents get task queues, memory and scheduling.

Works with

Model-agnostic: Anthropic, OpenAI, Google, xAI, DeepSeek, OpenRouter, and local models via llama.cpp. It also integrates with editors and IDEs — Claude Desktop, Cursor, Zed, and JetBrains — through MCP and ACP protocol support, and ships Tauri-based desktop apps for Linux, macOS and Windows.

How to install or connect

pipx install gptme

Alternatives shown in the README: uv tool install gptme or pip install gptme.

Maintenance and safety

MIT licensed, roughly 4.4k GitHub stars, 434 forks, and over 4,500 commits, with recent releases (v0.33.0, August 2026) adding sandboxed execution and new editing formats — an actively maintained project. Because it runs arbitrary shell and Python commands on your machine at the model's direction, it carries real risk if misused; the project mitigates this with a "lessons" system, pre-commit checks and structured task selectors for autonomous mode, but users should still add their own guardrails for unattended use.

Who should use it

Developers who want a scriptable, provider-agnostic terminal agent for day-to-day coding help, or who want to configure a supervised autonomous agent for repetitive tasks without being locked into one vendor's CLI. It's a good fit if you value the ability to switch model providers freely and don't mind a code-first, config-driven workflow over a polished GUI.