Test whether an LLM reliably renders the same UI correctly across multiple generations.
Add a CI gate that fails a pull request when generated UI accessibility or fidelity regresses.
Compare which LLM produces the most accurate and accessible UI using the built-in leaderboard.
| maryanprydatko/uivet | 0xkinno/neuralvault | 0xmayurrr/ai-contractauditor | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | hard | easy |
| Complexity | 3/5 | 4/5 | 2/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Offline demo needs no API key, live mode needs a Google or OpenAI API key.
uivet is a testing and evaluation tool built for teams whose product generates user interfaces using a large language model at request time. Normal visual regression tools assume the UI comes from deterministic code, so a fresh render that differs slightly every time defeats them, and normal LLM evaluation frameworks judge text answers rather than the pixels and structure of a rendered interface. uivet is built to sit in that gap: it generates the same scenario several times, renders each version in a real headless Chromium browser, and checks the actual rendered result rather than just the raw text the model produced. For each test scenario, uivet samples multiple generations, checks them against a set of deterministic rules such as fidelity to the required data, accessibility, layout correctness, and console errors, and can also send the render to a multimodal AI judge for a quality score. The results across the samples show how consistent a model's output really is, since one run silently dropping a piece of information, like a price, would otherwise go unnoticed. There is an offline demo mode that replays already-recorded generations with the AI judge turned off, so you can try it in about thirty seconds with no API key and no network calls. A live mode generates fresh UIs using a real model, defaulting to Google's Gemini, though it can also use OpenAI or any OpenAI-compatible server, and this mode does require an API key. uivet is designed to run inside continuous integration as a gate: a failing check or a detected regression against a saved baseline makes the run exit with a failure code, so a pull request that makes UI generation worse can be caught automatically. The project also publishes a public leaderboard comparing how different models perform on the same fixed set of scenarios, using only the deterministic checks with the AI judge turned off, to compare accuracy and accessibility across models. It is not yet published on npm, so it is installed directly from its GitHub repository.
A test and evaluation harness for LLM-generated UI that samples multiple generations, renders them in a real browser, and gates CI on fidelity, accessibility, and layout checks.
Mainly TypeScript. The stack also includes TypeScript, Bun, Playwright.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.