gptme/gptme is a personal AI agent that works inside a terminal. The project started on March 24, 2023, and it runs wherever a terminal runs: on a laptop, in an SSH session, under tmux, on headless servers and in CI pipelines. It carries shell command execution, Python code execution, web lookup and a vision tool that reads images, so the agent writes code itself and runs terminal commands on its own. The most recent release tag is v0.35.1.
The main value is that users can configure the agent themselves. The project is published so that people build and run continuously autonomous agents, it is provider-independent and it works local-first. It connects to model providers such as Anthropic, OpenAI, Google, DeepSeek and Ollama through subscriptions and local servers, so models can be swapped in. It is written in Python and the license is MIT. The repository star count at the time of the record was 4,457.
That said, the verifiable range is narrow. The material contains no change list and no performance measurements for v0.35.1, so what this version improved has to be checked in the repository release notes (https://github.com/gptme/gptme). The repository score records only 0.9 for momentum out of 10, and no freshness, technology or quality scores are given. The feature scope and the deployment environment are clear from the documentation, but there is no basis for saying that a particular version raised performance.