Overview
Vercel fx is an experimental coding agent open-sourced by Vercel Labs on August 17-18, 2026, under the Apache 2.0 license, hosted in the GitHub repository vercel-labs/fx. The tool is written in Zig and compiled into a single native binary, with a size of only 6.39 MiB, a cold start time as low as 10 microseconds, and a baseline memory footprint in the single-digit MB range. It has no JavaScript runtime dependencies, making it designed for resource-constrained environments such as CI containers, sandboxes, and low-spec VMs. fx is positioned as a lightweight, model-agnostic coding assistant, supporting multiple usage modes including an interactive CLI, a one-shot query mode, the Agent Client Protocol, and WebAssembly embedding, aiming to provide developers with a flexible and efficient automated coding experience.
Key Features
- Ultra-lightweight native binary: Written in Zig, compiled into a single native binary, only 6.39 MiB in size, with a cold start time of 10 microseconds, a baseline memory footprint in the single-digit MB range, and no JavaScript runtime dependencies, suitable for resource-constrained environments.
- Four usage modes: Provides an interactive CLI (Unix-like shell style), fx ask one-shot mode (supports --json structured output for scripting and CI), fx acp (Agent Client Protocol for editor-driven use), and WebAssembly with the libfx SDK (npm package, embeddable in browsers, with an online demo).
- Model-agnostic design: By default, connects to models via the Vercel AI Gateway, while also supporting OpenAI Codex subscription OAuth login (fx login codex) and xAI Grok subscription OAuth login (fx login grok), as well as local inference.
- Extensibility: Supports Skills, MCP servers, sub-agents, and session resumption, making it easy to integrate into complex workflows.
- Simple installation: One-line installation via curl -fsSL https://fx.sh/setup.sh | bash, with no additional dependencies.
Use Cases
- Run automated coding tasks in CI containers, leveraging fx ask's JSON output for scripted processing.
- Perform code generation and modification in low-spec VMs or sandbox environments, benefiting from extremely low memory and startup overhead.
- Integrate into code editors via the fx acp protocol for interactive coding assistance.
- Embed the libfx SDK in browsers to build online coding tools or demo environments.
- Use local inference models for offline coding assistance, protecting code privacy.
Pros
- Extremely low resource usage, suitable for edge or constrained environments.
- Model-agnostic, allowing flexible switching between different AI services.
- Multiple usage modes covering CLI, scripting, editor, and browser scenarios.
- Supports Skills and MCP, offering strong extensibility.
- Open source under the Apache 2.0 license, freely usable and contributable by the community.
Pricing
fx is an open-source tool, and the code is free to use. Model invocation costs depend on the chosen AI service (such as OpenAI Codex, xAI Grok, or local inference), with specific prices subject to each service's official website.
Summary
Vercel fx is an experimental lightweight coding agent written in Zig, with a small size, fast startup, and low memory usage, suitable for resource-constrained environments. It offers multiple usage modes, supports model-agnostic integration, and has extensibility features like Skills and MCP. Currently in the experimental stage, the official documentation notes that users assume the risk of use, but it has already gained community attention, receiving approximately 2,200 stars within five days of release.