Vercel CEO Releases xeet: X Post Previews in 1 ms Without a Browser, 100–1000x Faster
To see how a post looks on X, developers usually launch Chrome and take a screenshot, which is slow and resource-heavy. Vercel CEO Guillermo Rauch has released xeet, a Rust program that renders a post as a PNG in about 1 ms (one millisecond) without a browser. He says it is 100–1000x faster, and real Chrome checks its accuracy against tens of thousands of layout rules. The takeaway: agents can be trusted with complex work when there is an automatic reference to check against.
AI-processed from X: @rauchg; edited by Hamidun News
To see how a post will look on X, developers usually spin up a full browser: they launch Chrome, open the page, wait for it to render, and take a screenshot. This is slow and heavy, especially when hundreds of agents need previews at the same time. Guillermo Rauch, CEO of Vercel, has introduced xeet, a program written in safe Rust that renders a post as a PNG in roughly 1 ms (one thousandth of a second) and does not launch a browser at all. According to Rauch, this is 100–1000x faster than a Chrome screenshot, and the output is just as accurate. For developers, this means a cheap preview of a post directly inside code or an agent.
What xeet does
xeet is an X post renderer written in Rust and compiled to WebAssembly, a format that runs almost anywhere. The author calls it hermetic: it contains no browser, no network access, no filesystem access, and no system fonts. It takes a post description as input and outputs a PNG.
A single post renders in about 1 ms, and in one example in 0.4 ms. The speed and the claim of "browser-level accuracy" come from the author; the post includes no independent measurements.
To match the browser's output, xeet reproduces what Chrome normally does:
- text shaping, as in the Blink engine and the HarfBuzz library;
- line breaking accurate to 1/64 of a pixel;
- bidirectional text, such as Arabic mixed with Latin (RTL, bidi);
- 28 subsets of the Chirp font that X uses;
- fallback fonts for Chinese, Japanese, and Korean (CJK), math symbols, and emoji;
- the "Show more" button rules, meaning exactly where a long post gets truncated.
Rauch himself admits that a human engineer would hardly manage this work by hand: there are too many small rules, and each one has to be checked against a real browser.
How it is built: the "autorepo"
Rauch calls the project structure an "autorepo" and says it was built around "verification engineering" from the start. The setup works like this:
- Oracle. It runs real Chrome and checks tens of thousands of text layout rules. The browser is the reference: if xeet renders something differently, that is a bug.
- CI watcher. An automated check in CI tracks how the original changes, meaning how posts display on X itself.
- Signed PRs for agents. When a discrepancy is found, the watcher opens a signed pull request, and an AI agent takes it from there.
In everyday terms, this is a school test with an answer key. The student (the agent) can write as many solution attempts as it likes, but a person does not grade them; the key (Chrome) does. As long as the key is unambiguous and there are plenty of checks, the agent can be given large, tedious work.
What this means in practice
First, xeet has a clear use. If you are building an agent that writes posts, or a scheduled-publishing service, checking how a post looks in a millisecond costs far less than maintaining a pool of browsers. Rauch stresses that previews help both people and agents: an agent can verify on its own that its text will not get cut off by "Show more" and that the layout holds.
Second, the takeaway goes beyond this project. A task that looks unmanageable for a human becomes achievable when it has a reliable automatic reference. You build the oracle and the checks first, then hand the implementation to agents. The result depends not on how clever the agent is, but on how well you can verify its work.
Caveats
The speed figures and the claim of "browser-level accuracy" come from the author, with no public benchmark in this post. There is also no information about the license, how often it is updated, or how the tool will hold up after X changes its design or truncation rules. An approach where the reference and the watcher are tied to someone else's changing platform requires ongoing maintenance: discrepancies will keep appearing, and agents will have to fix them.
Still, xeet shows a working pattern: a small, fast component, a strict reference, and agents that adjust code to match it. If you have a narrow but highly detailed task, such as rendering, parsing, or compatibility with someone else's format, this setup is worth trying.
- Nov 21, 2026VibeCBS collects project nicknames in advance: the agent finds 2.8 times more documents
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026Sakana AI Opens Applied Research Lab for Market and Economic Modeling
- Nov 21, 2026Meta Is Reportedly Preparing the $300 Muse Charm AI Tamagotchi With a Year of Verizon 5G
Key Questions Answered
What xeet does?
xeet is an X post renderer written in Rust and compiled to WebAssembly, a format that runs almost anywhere. The author calls it hermetic: it contains no browser, no network access, no filesystem access, and no system fonts. It takes a post description as input and outputs a PNG. A single post renders in about 1 ms, and in one example in 0.4 ms. The speed and the claim of "browser-level accuracy" come from the author; the post includes no independent measurements.
How it is built: the "autorepo"?
Rauch calls the project structure an "autorepo" and says it was built around "verification engineering" from the start. The setup works like this: Oracle. It runs real Chrome and checks tens of thousands of text layout rules. The browser is the reference: if xeet renders something differently, that is a bug. CI watcher. An automated check in CI tracks how the original changes, meaning how posts display on X itself. Signed PRs for agents.
What this means in practice?
First, xeet has a clear use. If you are building an agent that writes posts, or a scheduled-publishing service, checking how a post looks in a millisecond costs far less than maintaining a pool of browsers. Rauch stresses that previews help both people and agents: an agent can verify on its own that its text will not get cut off by "Show more" and that the layout holds. Second, the takeaway goes beyond this project.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.