Skip to content

Discover

Repositories gaining contributors quickly that are still small enough to be worth telling someone about. Anything over 2K stars is excluded outright, along with forks and archived projects — the point is what you have not seen, not what is merely busy.

Growth is measured in distinct contributors, never stars: this archive stopped recording stars during 2026, so a star-based ranking would describe the archive rather than GitHub. Star counts here come from GitHub's own API and are used only to rule the famous out. Categories are read from the repository description and name.

Partial coverage10,540 / 10,587 hourly files (100%) · 2 absent upstream · 44 failed, retryable2025-05-262026-08-10 (UTC)— sampled evenly across the window, so rankings and trends hold; absolute counts scale up.

144 repositories across 12 categories, from a pool of 1,359 under the star ceiling. At most 2 per owner in a category, which is what keeps a single account from filling a section.

AI agents & skills

Agent harnesses, skills, MCP servers and the tooling built around coding assistants.

Models & inference

Running, serving, quantising and fine-tuning models.

Extraction & parsing

Turning documents, pages and files into structured text.

Infrastructure

Deployment, containers, self-hosting and the plumbing underneath.

Developer tooling

CLIs, editors, SDKs and libraries aimed at people who write code.

Web & interfaces

Front ends, APIs and things with a browser on one end.

Data & databases

Storage, query engines, pipelines and analytics.

Security

Offensive and defensive tooling, OSINT and reverse engineering.

Media & games

Video, audio, graphics and interactive work.

Mobile

iOS, Android and cross-platform application work.

Learning & collections

Courses, guides and curated lists.

Everything else

Growing fast, but the description does not say enough to place it. About a third of the pool lands here — these are worth a look precisely because no rule caught them.

Read this as a shortlist, not a ranking. Categories come from keyword rules over a description that reaches the archive through fork events — so it is occasionally a forker's edit rather than the project's own, and roughly a third of the pool matches no rule at all and sits under “Everything else”. The spam filter catches the obvious auto-generated repositories; it will not catch a determined one.