Vercel Labs Open-Sources fx: A Zig Coding Agent Designed to Be Embedded

Vercel Labs released fx as open source, an experimental coding agent and command-line tool written in Zig that ships as a 6.3 MiB native binary with a reported 10-microsecond cold start and single-digit-megabyte memory baseline, designed to be scripted or embedded rather than run as a full-screen terminal app. The tool exposes the same engine through a CLI and the libfx library, including a WebAssembly build, supports structured JSON output for automation, and lets editors attach over the Agent Client Protocol via `fx acp`; release 0.0.11 added named OpenAI-compatible connections for local servers and alternative gateways, and release 0.0.13 on October 4 added an Ultrafast mode on supported OpenAI models while warning that new libfx checkpoints cannot be restored by older versions. Vercel states fx is experimental and pre-1.0, and a GitHub issue flagged that its name collides with the long-established terminal JSON viewer fx, which already holds the Homebrew `fx` formula.

Vercel Labs has released fx as open source https://x.com/vercel dev/status/2089828083415355806 , an experimental coding agent and command-line tool written in Zig and built to be scripted or embedded rather than run as a full-screen terminal application. Announced in August 2026, fx ships as a 6.3 MiB native binary with a reported single-digit-megabyte memory baseline and a 10-microsecond cold start, figures Vercel states rather than independently verified benchmarks. Instead of taking over the terminal with a persistent interface, fx writes its conversation into ordinary scrollback and exposes the same engine through both a CLI and a library. The result is aimed at developers building agent infrastructure more than at people looking for another chat window. Structured output is available for automation, so a script or CI job can call a command and parse the response: fx ask --json "summarise the failing tests" Editors can attach over the Agent Client Protocol using fx acp , and libfx https://github.com/vercel-labs/fx exposes the engine to JavaScript, including a WebAssembly build. Early Hacker News feedback https://news.ycombinator.com/item?id=49353339 also probed the model-agnostic claim. Some first-run reports said setup appeared to route inference through Vercel AI Gateway, and commenters questioned how fx differed from established agents, pointing to OpenCode's more polished terminal experience. One user noted GLM 5.2 being free: Just tried and although it’s not that polished or fast imo it’s very cool. GLM 5.2 totally free even with a free Vercel account. Make business sense too - fx is the entry point to bring more user to Vercel AI. Provider choice has since widened: release 0.0.11 https://fx.sh/changelog.md added named OpenAI-compatible connections for local servers and alternative gateways, and the README now lists Vercel AI Gateway, ChatGPT and Grok subscriptions, plus endpoints such as Ollama and OpenRouter. Vercel has also slotted fx into its AI SDK harness layer https://vercel.com/changelog/fx-ai-sdk-harness-adapter next to Claude Code, Cline, Codex, Cursor, OpenCode and Pi, presenting it as one option behind a shared interface. A GitHub issue https://github.com/vercel-labs/fx/issues/197 flagged that the name collides with the long-established terminal JSON viewer fx, which already holds the Homebrew fx formula, and a contributor confirmed that packaging and command coexistence were being reviewed. Vercel is explicit that fx is experimental. The 0.0.13 release https://github.com/vercel-labs/fx/releases/tag/v0.0.13 on October 4 added an Ultrafast mode on supported OpenAI models with quicker launches and shell calls, but warned that newly written libfx checkpoints cannot be restored by older versions. CLI users can move up with fx upgrade and should read the changelog https://fx.sh/changelog.md first, while embedded users should consult the libfx migration guide https://fx.sh/docs/lib/node.md migrating-from-0013 , which explains that the earlier createFxAgent call maps to createFxEngine in the current documentation https://fx.sh/docs/lib . Its repository https://github.com/vercel-labs/fx documents the CLI, skills, MCP, permissions and SDK, and its browser demo https://fx.sh/try shows the terminal-style experience without an install. fx is an open-source project from Vercel Labs, the experimental group within Vercel, the company behind the Next.js framework and its platform for building and deploying web applications. Written in Zig and distributed as a single native binary, fx is designed to be run from the command line, scripted in automation, or embedded directly in other applications through its libfx library and WebAssembly build. Still pre-1.0 and changing quickly, it reflects a wider move toward treating coding agents as composable infrastructure rather than standalone tools.