Radar

Star velocity over 7 days, ranked by acceleration. Computed 2026-08-27.

4,722 of 5,526 sampled repositories have enough history to measure a rate.

Or compare two, or start from what you already have — paste your package.json. It is read in your browser and never sent.

• react540
• react-dom511
• pydantic431
• numpy384
• zod359
• requests350
• pyyaml313
• fastapi308
• python-dotenv297
• httpx292
• next292
• uvicorn286
• clsx271
• lucide-react265
• openai263
• tailwind-merge239
• rich237
• pandas219
• class-variance-authority209
• pillow209
• tqdm207
• torch176
• transformers171
• click153

The fastest 100 of 4,722 repositories with a measurable rate. Every one of them entered the sample within the last 30 days, which is a fact about how recently riivault began watching and not about how old any of these repositories are.

What this sample is. Public repositories found by GitHub Search: topic queries (AI topics first) and language sweeps, each banded by star count because the API returns at most 1,000 results per query. Forks, archived repositories and anything under the star floor are excluded. Topic queries can only reach repositories whose authors added topics, which over-represents templates and starters; the language sweeps exist to reach the rest. This is what was measured, not a census of what developers build.

A star is attention, not adoption. Nothing on this page says anyone shipped these to production.

The caveats are reported separately because they are different claims. Discussed means a Hacker News thread inside the window explains the climb — riivault can name the cause. Not being pushed means stars are arriving at a repository nobody is working on, which is the shape published estimates put millions of purchased stars in. Younger than the window is not about the climb at all: the repository was created inside the period being measured, so its rate averages its whole life rather than measuring one period against an earlier one, and the first reading here is already mid-climb. It is how a row can be arithmetically exact and still not say what it appears to. None is proof of anything, and none excludes a row.