Where GitHub is moving

23 topics, two lenses. Attention: new stars every day on 3.2k of the biggest and newest repos in each topic. Building: how many new repos people create with each topic's tags. In the last 30 days MCP servers (+17% vs field), Image generation (+14% vs field), Learning and lists (+54% vs field) pulled ahead, while Prompt libraries and Self-hosted apps cooled.

Share on X

Window

Priority for

What to hop on,

Heat: attention against the field (55%), more new repos (30%) and room at the top (15%, no single repo taking the stars). Jev's priority (TypeSafe Jev, 0 to 4 per role) reads the same numbers.

Velocity board

Sort

Heat map, one year

Lens

Rows sorted by heat for the chosen window. Each cell is the topic's share of all the new stars we track that week. "Own peak" shows each topic against its own best week, so a row that turns white toward the right is taking a bigger slice than before. "Share of all" puts every row on one scale, so it shows which topics are big. Building is by month; the dashed last cell is the month so far.

Big and rising, or small and fading

Right: a bigger share of all attention. Up: growing faster than the field in the chosen window. Bubble size: new repos created in the window. Filled: hop on now.

Does the 80/20 rule hold?

Last 30 days. Bar: share of new stars that go to the top 20% of repos in each topic. White tick: the single biggest repo. Dashed line: 80%. Past the line, a few repos take almost everything; well short of it, the field is open.

Compare two topics

Check first pressure Open the Trust check

Share of checked repos in each topic whose stars or install path need a closer look.

How this is measured

Attention: daily new stars from GH Archive (via OSS Insight) for 3.2k repos found by searching each topic's tags for the 50 most-starred active repos and the 25 most-starred newcomers. A repo in several topics is split evenly between them, so shares add up to 100%. Because the sample is today's leaders and newcomers, every topic looks like it grew over long windows; "vs field" divides that out, so use it to compare topics.

Building: GitHub Search counts of new public repos (forks excluded) created with each topic's tags, per day for 60 days and per month for a year. Tags are often added days after a repo is created, so the newest days are counted again until they are two weeks old; the latest window can read a little low. GitHub as a whole creates far more repos than a year ago (all 23 topics together: 55% in 30 days, 196% over 12 months), so new repos are shown against that field, like attention.

Check first: 16 repos whose stars match the radar's patterns (held by the radar, over 60 days old with 80%+ of all stars in 30 days, or 3+ same-owner repos doing that) are left out of every number. Patterns fit launch pushes and bought stars alike.

Refresh: daily. Star history is fetched across several machines because OSS Insight allows 600 calls an hour per address.