FuXi Freemium

What it does

FuXi is a terminal-based, provider-agnostic AI coding agent — pitched as an alternative to Claude Code that works with any OpenAI-compatible model, plus Gemini and Bedrock/Vertex. It runs a "Think → Act → Verify" loop: reading and editing code, running shell commands, and verifying its own changes, with cost-aware routing and automatic failover across whichever providers you configure.

What is inside

50+ built-in tools (file read/write/edit, bash/PowerShell, ripgrep search, web fetch, LSP diagnostics, Jupyter, browser automation, background tasks, parallel sub-agents), an MCP client, hooks/skills/plugins, durable on-disk sessions with checkpoints and rollback, and a self-updating single static Go binary with no runtime dependencies.

Works with

Any OpenAI-compatible API endpoint, Gemini, and Bedrock/Vertex; extensible via MCP servers.

Pricing

Published at fuxicode.com/en/pricing as a tiered, usage-based model. Free ($0/month) covers the core terminal agent, code generation, file operations, session memory and MCP/plugin extensibility. Hobby ($0/month, marked "Recommended") adds everything in Free plus a one-time $15 usage credit on signup and more multi-model usage-based capacity. Pro and Ultra are both listed as "coming soon," promising higher reserved usage limits, more models and priority access at busy times; FuXi auto-routes each task across three internal capability tiers (Lightweight, Standard, High Performance) to balance quality against cost, and exact per-token rates only show up after you log in.

How to install or connect

curl -fsSL https://downloads.fuxicode.com/bootstrap.sh | bash

Windows (PowerShell): irm https://downloads.fuxicode.com/bootstrap.ps1 | iex. Verify with fuxi doctor.

Maintenance and safety

Actively maintained into September 2026, with an AST-based shell-command safety classifier, fine-grained permissions, audit logging, and a documented security and privacy program. The source code is proprietary — the GitHub repo contains only documentation, installers and the issue tracker — and an optional flag exists to bypass all safety checks, which the docs themselves label dangerous.

Who should use it

Developers who want a Claude-Code-like terminal agent but need to bring their own model or provider rather than being locked to one vendor, and who are comfortable using a free-to-use but closed-source binary rather than an open-source tool. It particularly suits teams that already have OpenAI-compatible, Gemini or Bedrock/Vertex access and want cost-aware routing and automatic failover across those providers baked into the agent loop, rather than hand-rolling that logic themselves, and who value the published self-run benchmark comparison over having to take performance claims on faith. Weigh the closed-source nature of the binary, and the free $15 signup credit's limited runway, against that flexibility before adopting it for anything security-sensitive or budget-critical.