A READER’S GUIDE
How to read the Agent Index
Use aiagent.club to find projects worth a closer look—not as a final verdict. Index v2 ranks only projects with current evidence across multiple source types, so one large or outdated number cannot tell the whole story.
Best for
Building a shortlist, spotting breakout projects, and comparing traction on the same observable signals.
Read together
Look at the score, its three components, source coverage, freshness, and recent change—not the rank alone.
Not a verdict
A high score means stronger relative traction. It does not certify code quality, security, or fit for your use case.
What goes into the score
Every software project is evaluated on three dimensions. An observation counts only when it is no more than two days behind the latest scoring snapshot; stale and missing values do not enter the score.
Observed adoption
Weekly npm downloads and monthly PyPI downloads, normalized to a monthly scale for comparison.
Useful as an adoption proxy; it can include CI, mirrors, caches, and automated traffic.Momentum
The project’s trailing seven-day GitHub star gain, measured from the daily time series.
Shows what is accelerating now rather than rewarding only projects that started early.Attention
The project’s current GitHub star count, used as a public signal of awareness and interest.
Stars measure attention, not active users, production deployments, or satisfaction.What one “project” represents
A project page may combine a GitHub repository with its npm or PyPI packages when source metadata identifies them as the same product. Multiple packages from the same registry use the largest current adoption value rather than being added together. Source sections show the records that support the score and link to their metrics and trend.
Category labels are inferred from public names and descriptions. A project may appear in several categories, and categories do not affect its score.
A practical way to use the site
- 1
Discover
Start with the overall index or a category to find a manageable set of candidates.
- 2
Compare
Open the project pages and compare adoption, momentum, attention, and source coverage side by side.
- 3
Verify
Follow a source record to inspect its current value and history; check Data status for freshness.
- 4
Follow
Add candidates to your Watchlist or use the weekly feed to catch changes instead of rechecking every day.
INDEPENDENT BY DESIGN
What keeps the ranking independent
Projects do not need to be claimed or paid for to be ranked. Owners cannot edit their score, and sponsorship or self-reported claims are not inputs to the formula. The score is calculated only from public records already tracked by the site.
Collection is scheduled twice daily, while each metric keeps one value per day. A partial run does not invalidate records that were collected successfully; a missed record keeps its history but ages out of the score after two days. Partial and failed runs are shown on Data status. Same-day reruns, documented backfills, identity merges, or corrections may update a record.
Limits you should keep in mind
- Observed adoption is not the same as active users, revenue, retention, or production usage.
- Download windows and platform definitions differ; figures from different sources should not be treated as identical measurements.
- OpenRouter throughput reflects activity on OpenRouter, not the entire model market.
- Newly discovered or single-source projects remain in the Observation pool until enough current, cross-source evidence is available.
- Automated categories can be imperfect. Treat them as discovery filters, then check the project itself.
Want to inspect the evidence?
The data page shows source coverage, freshness, and collector health. Every project page links back to its source records, and the collectors are available in the public repository.