Fuxi

Why we included this project

The pitch is simple: run an AI coding agent from the command line instead of a chat sidebar. FuXi's Think-Act-Verify loop reads code, edits files, and runs shell commands, then inspects the result and repeats until a failing test is fixed or the pull request is ready. Every request is scored by complexity and routed to the right model tier across multiple providers, so simple tasks go to the cheap models and the hard ones use the stronger models, which keeps the cost per job reasonable. It ships as a single static binary with no runtime dependencies, works from a keyboard-only TUI, acts as an MCP client, bundles more than fifty built-in tools, applies safety rules around what the shell may run, and keeps session transcripts you can resume or roll back, so long autonomous sessions do not have to start from zero. One honest caveat: the download is a free binary under a proprietary license, not open-source code, so do not expect to read or modify the agent itself.