Main

Treat gptme as a “tool runner with a chat UI”:

• Put it in a repo and feed it structured input (git diff, make test, logs) so it can act on concrete artifacts.
• Prefer -n in CI so it never blocks for confirmation; use -y for supervised runs you can interrupt.
• If you need web access, install the optional extra (gptme[browser]) and validate the sandboxing policy for your environment.

README excerpt (verbatim)

gptme

/ʤiː piː tiː miː/
_{what does it stand for?}

Getting Started • Downloads • Website • Documentation

📜 A personal AI agent that runs anywhere a terminal runs — your laptop, ssh sessions, tmux, headless servers, CI pipelines.
Provider-agnostic, local-first, and unconstrained: ships with shell, Python, web, vision, and everything else an agent needs.
A great coding agent, but general-purpose enough to assist in all kinds of knowledge-work.

Free and open-source. Works with Anthropic, OpenAI, Google, xAI, DeepSeek, OpenRouter, or fully local via llama.cpp — your data, your models, your terminal.

FAQ

Q: Does gptme include MCP support? A: README states MCP support is included by default; you can add extras for more capabilities.

Q: What Python version is required? A: Prerequisites list Python 3.10 or newer.

Q: How do I make it safe for CI? A: Use -n for fully non-interactive runs, and consider -y only when a human can monitor and interrupt.