CheetahClaws
A fast agent harness for long-running, multi-model, tool-using AI systems, installed as a Python package that starts a chat-style terminal session and can also run inside a bundled desktop app.
CheetahClaws is an agent harness infrastructure for long-horizon, multi-model, tool-using AI systems, installed with a single pip command and started by running its own command to begin chatting immediately with a chosen model, such as a DeepSeek model. A separate desktop build wraps the same interface in a window for people who prefer not to use a terminal.
Recent releases focus on reducing interruptions during a session: requests are approved automatically by default rather than stopping at a menu for routine actions, while a hard denylist, a filesystem sandbox, and manual or plan modes still take precedence over that autonomy; permission prompts are now reserved for actions that can change files, run arbitrary code, or reach outside the session. Sessions can be resumed by name from a list showing each one's opening request, date, turn count and model, and are saved as readable, greppable UTF-8 text rather than escaped characters.
It is written in Python and distributed in several languages.