Research follow-up

100 open-source projects, 51 days later: where attention moved

A fixed-cohort refresh of our September research: 183,541 additional GitHub stars, category changes, activity checks and the limits of licence metadata.

19 November 2026Independent research

This research continues The 100 open-source projects that matter in 2026 (September baseline) and What we found looking at 300+ cloud marketplace listings (August fieldwork). Those articles establish the cohort and the commercial questions; the evidence below states what we checked again.

What changed in 51 days

We queried the same 100 GitHub repositories on 19 November and compared their current totals with the figures published on 29 September. The cohort gained 183,541 net stars, from 7,278,500 to 7,462,041: a 2.52% increase. The 22 AI repositories accounted for 98,860 of those additional stars, or 53.9% of the total change.

This is a repeat measurement of an editorial selection, not a ranking of all open source. Holding the cohort fixed matters: adding this month's popular projects would mix changes in attention with changes in our selection.

The category picture

CategoryProjectsSeptember starsNovember starsNet changeChange
AI, LLM and agents222,830,6902,929,550+98,8603.49%
Cloud and containers13496,109503,365+7,2561.46%
Databases, search and streaming15519,834527,513+7,6791.48%
Observability9341,997346,639+4,6421.36%
Security and identity9216,593223,389+6,7963.14%
Developer tools and runtimes131,057,7231,079,071+21,3482.02%
Web and application platforms111,108,9011,124,228+15,3271.38%
Self-hosted and local-first8706,653728,286+21,6333.06%

AI had the largest absolute increase and the highest percentage increase among these eight groups, at 3.49%. Security and identity grew 3.14%, and self-hosted software grew 3.06%. These are category totals: one large repository can drive a group's result. They are not the average growth rate of a project in that category.

Where the increase concentrated

ProjectSeptemberNovemberNet changeChange
opencode210,745227,984+17,2398.18%
ComfyUI135,459147,638+12,1798.99%
Codex CLI127,067135,876+8,8096.93%
llama.cpp129,847135,708+5,8614.51%
Ollama181,904187,621+5,7173.14%
Open WebUI153,511159,087+5,5763.63%
RustDesk124,764130,248+5,4844.40%
n8n206,247211,399+5,1522.50%
Browser Use116,684121,488+4,8044.12%
Immich115,268119,648+4,3803.80%

opencode, ComfyUI and Codex CLI together contributed 38,227 net stars, 20.8% of the cohort's increase. ComfyUI also led percentage growth within the fixed cohort, at 8.99%. A large existing audience and a high growth rate are different observations; this table makes both visible.

Activity and licensing: what this refresh can establish

All 100 repository requests succeeded. None returned an archived or disabled flag, and every reported push timestamp fell within the preceding eight weeks. No requested owner/repository path resolved to a different canonical name, ignoring letter case. These checks describe repository metadata. A recent push does not establish a supported release, maintainer responsiveness, secure dependencies or a working upgrade path.

GitHub returned NOASSERTION for the licence identifier on 21 repositories. That is an inconclusive automated classification, not a finding that 21 projects lack licences. GitHub's licence API documentation explains that its detector matches licence files against known licences and does not account for dependency licences or every way a project documents its terms.

We therefore preserve September's manually reviewed licence labels as historical observations. This refresh does not re-certify those labels. Before distributing or hosting a particular release, inspect that release's licence files, commercial subdirectories and dependencies. The metadata field alone cannot answer that question.

What the numbers do not tell us

Stars are cumulative bookmarks. The difference between two totals is a net change, including unstars; it is not a count of new users, deployments or paying customers. The same person can star multiple projects, so summing stars does not count unique developers. This sample also excludes projects that emerged after our September selection.

The earlier marketplace investigation asked whether packaging widely used software produces a viable business. This refresh cannot answer that. Attention can identify candidates for interviews; willingness to pay, support costs and renewal behaviour require separate evidence. Our companion marketplace economics note tests the fee arithmetic without treating popularity as demand.

Method and data

We extracted the numbered rows from the September article and queried GitHub's repository endpoint once per repository. We retained repository IDs, canonical names, counts, flags and timestamps. Requests completed within a short collection window, not at one atomic instant. The September baseline is the published table; a separate archived September API response was not available in this checkout.

Download the complete 100-project comparison (CSV). The file includes both dates' counts, net changes, activity timestamps and API licence classifications. It supports recalculating every aggregate in this article. It does not contain a historical star-event series or a new licence audit.

Sources checked on 19 November 2026. Observations, calculations and limitations are identified separately. Corrections are welcome.