fx is a minimal coding agent harness and CLI from Vercel Labs. It is a single native binary written in Zig (about 6.7 MiB for v0.0.14), so there is no runtime to install, and it runs the same on a laptop, in CI or inside an agent sandbox. Its output prints into the terminal scrollback like a normal command rather than taking over the screen. The project is open source under Apache 2.0 and is explicitly labelled experimental by Vercel Labs.
fx reads code, edits files and runs commands from an interactive session, or answers a single request with fx ask (with --json output for scripts). Editors connect through the Agent Client Protocol (fx acp), skills are loaded from .fx/skills as well as .claude/skills and .agents/skills, MCP servers are managed with fx mcp, and projects can add instructions through AGENTS.md. libfx exposes the same agent core as a JavaScript library (native Node.js addon and WebAssembly); v0.0.14 (2026-10-16) made libfx sessions durable across crashes and redeploys.
Model access is bring-your-own: sign in to Vercel AI Gateway, reuse an eligible ChatGPT (Codex) or Grok subscription via OAuth, or point fx at any OpenAI-compatible Chat Completions endpoint such as Ollama or OpenRouter. Conversations and settings are stored locally under ~/.fx; the docs state that fx does not automatically upload the workspace, Git history or session files, but prompt context goes to the selected provider.
fx runs on macOS and Linux (x86_64 and arm64) and can be tried in the browser via WebAssembly. There is no fx-specific subscription; Vercel's separate Token Pass (early access for Pro teams, $30/month for $80 of GLM 5.3 and Kimi K3 credit) is one optional way to pay for tokens.
