Methodology

Every number on What's Trending in AI is computed from collected data on a fixed schedule. Nothing is editorial, and rankings cannot be bought.

1. Daily data collection

Every day we snapshot 674 open-source projects across four categories (MCP servers, AI agents, agent frameworks, agent SDKs): GitHub stars, forks, contributors, open issues, releases, and last-push time, plus weekly npm and PyPI download counts where a package exists. Snapshots are append-only — once written, a day's numbers are never edited. That immutable history is what makes week-over-week claims verifiable.

2. Weekly growth score (0–100)

Each week, every project gets a composite score from four components:

Acceleration — 40%
Stars gained this week vs the previous two weeks. Measures whether growth is speeding up, not how big the project is. Tiny absolute numbers are gated out so noise never scores.
Relative velocity — 30%
Stars gained as a percentage of total stars — a 500-star project gaining 100 stars outpaces a 50,000-star project gaining 200.
Downloads — 20%
Weekly npm + PyPI downloads on a log scale, discounted when star momentum is low so install-base alone can't carry a stagnant project.
Activity — 10%
Recency of the last push: pushed today scores 100, untouched for 30+ days scores 0.

3. Signals

🔥 Trending
Star growth at least doubled week-over-week, or a sustained spike of 500+ stars in a week — with a growth score of 70+. Details on the trending page.
🌱 Emerging
Young (≤ ~6 months), still small (< 5,000 stars), already gaining 30+ stars a week. Details on the emerging page.
📉 Declining
An established project (500+ stars) running below 20% of its own 8-week baseline velocity. Measured against itself, never against others. Details on the declining page.
♻️ Deprecated
Officially replaced by a successor. A lifecycle state, not a signal — a deprecated project never appears in trending, emerging, or declining, and a successor's early growth is treated as migration, not emergence.

4. Cold start & honesty rules

Data as of 2026-06-10 · scores for week of · methodology version: B0 (7-fix revision, validated 2026-06).