Fuxi: An AI coding agent that runs in the terminal as a single Go binary

This is an agent that directly reads code, edits files, and executes commands within the terminal. Written in Go, it is distributed as a single static binary with no runtime dependencies, allowing it to run immediately after installation. It provides a workflow similar to Claude Code, but without vendor lock-in, enabling free connection to OpenAI-compatible models.

Through a Think → Act → Verify loop, it goes beyond simple answer generation to work on actual codebases and verify results. It features over 50 built-in tools, including file I/O, shell execution, ripgrep search, LSP diagnostics, and web fetching. Before executing shell commands, it passes through an AST-based safety classifier and offers fine-grained permission management and audit logs to control the risks of autonomous tasks.

It supports cost-aware routing between LLM providers and automatic failover. Conversation history is persistently stored on disk, and checkpoints allow for session resumption, rollback, and forking. During idle states, a 'dream' function consolidates memory, and long conversations are automatically compressed to save tokens. It can be extended via hot-reload for MCP clients, hooks, skills, plugins, and more.

It is free for individuals, teams, and enterprises with no license fees. It provides self-update functionality with background version checks and checksum verification. Rather than publishing benchmark scores, it emphasizes transparency through reproducible self-evaluation and real-time monitoring commands such as /cost and /usage.