How it works

A curated set of open-source AI projects: the frameworks, model runtimes, agents, retrieval tools, and infrastructure being built in the open. The goal is to help you see what is being built and what is gaining momentum, including smaller projects that could be relevant to your work.

Projects are curated, not crawled indiscriminately. Each tracked repository is added deliberately, so the set stays signal-rich rather than exhaustive.

Every tracked project gets a daily snapshot of public GitHub metrics: stars, forks, open issues and pull requests, merged pull requests and commits over the last 30 days, the latest release, and the last push. Data is read only during that scheduled sync, never when you load a page.

When a metric is genuinely unavailable, it is stored as null, never as a fabricated zero. A brand-new project shows as “newly tracked” until it has enough history. It never gets a fake score.

Ranking uses a transparent score, momentum-v1. It turns six growth and activity signals (star and fork growth over 7 and 30 days, merged pull requests, and recency of activity) into percentiles across the tracked cohort, weights them, and renormalizes by whatever data is actually available, so scores stay comparable and nothing is invented.

The full formula, weights, and eligibility rules are public and always match the code.

Read the full methodology →

Two curated facets sit alongside the metrics:

• Category. What the project does (LLM frameworks, model serving, agents, retrieval & RAG, local AI, and more).
• Affiliation. Who is behind it: Independent, Startup, AI Lab, Institutional (academic), or Big Tech. Both are shown as badges and can be filtered on the explorer.

These are human-assigned and shown openly. They are curation, not a hidden algorithm.

The core principle: no black boxes. Rankings come from public metrics with published weights; categories and affiliations are labeled human curation. Synchronization only ever updates GitHub-controlled fields and never overwrites human-authored content. You can always answer “why does this project rank here?”

Once a month, an issue is rendered from this same real data: the projects gaining the most momentum, with the numbers behind each pick. It is never fabricated, and it only goes to people who subscribe on this site. Sign up from the home page or the footer.

The site is public and read-only: no accounts, no logins, no ads. It does not rank projects by opaque popularity or by anyone paying for placement. It is intentionally small and focused on being useful and honest over being exhaustive.