aiminute. ← All AI news
Tools 2026-08-20

A coding agent that is six megabytes and starts in ten microseconds — Vercel Labs wrote it in Zig and gave it away

A coding agent that is six megabytes and starts in ten microseconds — Vercel Labs wrote it in Zig and gave it away

Vercel Labs has open-sourced fx, a command-line coding agent written in Zig and released under Apache-2.0. The pitch is size and speed: a single binary of about 6.4 megabytes that cold starts in roughly ten microseconds and, by design, does no unnecessary work or input/output before it accepts your first instruction. It is deliberately model-agnostic — it will talk to a local model, a gateway, a provider API directly, or a subscription you already pay for — and it follows Unix conventions rather than wrapping everything in its own environment, so it composes with the shell tools already on the machine. The project is at version 0.0.4 and its own documentation labels it experimental. It drew a few hundred points of discussion on Hacker News after release.

Why it mattersAlmost every coding agent shipping this year is a large runtime that arrives with its own model relationship, its own configuration surface and, increasingly, its own opinions about where your code should live. fx is an argument that the agent could just be a small program. The concrete consequences are that it starts fast enough to invoke from a shell loop or a git hook without thinking about it, and that swapping the model behind it is a configuration change rather than a migration. Ten microseconds is not a number that matters when you type a command once, but it matters when a script calls it a thousand times. The caveat is the version number: 0.0.4 and self-described as experimental means this is a design worth looking at rather than a tool to depend on yet. What is interesting is the shape, not the release.
#Coding#AI Agents

✓ Verified · 3 sources

WhatsApp X Telegram
Read in the app — free, in 9 languages

Related stories

Apple Music will tell you when a song was made by a machine — but the uploader decides whether to say so
2026-08-21
Rumour: the anonymous model that just topped a coding benchmark, for free, is said to be Zhipu's unreleased flagship
2026-08-21
DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best
2026-08-21
Nvidia is paying $6 billion for the machine that builds a rival's models — and hiring 109 of the people who ran it
2026-08-21
Stripe has just paid $7.5 billion for a model router. Days later Ramp built one and is giving it away until January.
2026-08-21