Open Source Says No went live as a searchable catalog of repositories that publicly reject LLM-generated contributions — refreshed November 11, 2026 with 22 verified entries and a community suggest workflow. The site is not a standards body. It is a snapshot of projects from Immich and Meilisearch to QEMU, Servo, and Gentoo that already spelled out bans in contributing docs, issue templates, or checklists.
The list landed on Hacker News the same morning, where maintainers, agents users, and skeptics argued volume, disclosure, and whether a directory helps or paints a misleading picture of all open source.
TL;DR
| Question | Answer |
|---|---|
| What changed? | New catalog at osssaysno.com tracking explicit AI bans |
| Count (Nov 11, 2026) | 22 verified projects |
| Largest on the list | Immich (~118k GitHub stars) |
| How entries are added | Suggest URL → prefilled issue → automated policy scan |
| Notable names | Meilisearch, OkHttp fork, Servo, QEMU, Veloren, MonoGame |
| Relation to prior bans | Many entries match policies explainx.ai already covered individually |
| Author stance (HN) | Early experiment; no claim that OSS universally says no |
What the catalog actually tracks
Each card on Open Source Says No includes:
- Repository name and short description from GitHub
- Quoted policy line pulled from contributing docs or templates
- Platform (mostly GitHub; Gentoo is listed as self-hosted)
- Scope tags such as code, docs, issues, art, or translations
- Star count and verified date (for example Immich verified 2026-11-11)
The homepage groups filters by category and platform. Hacker News commenter lemagedurage pointed out that clicking All projects shows many rows tagged restricted — a UI label for stricter policies, not proof that most of open source bans AI. The author jlengrand replied that the domain name is tongue-in-cheek and that they do not intend broad conclusions from a partial sample.
That honesty matters for readers comparing this catalog to COSMIC’s blanket ban, QGIS’s AI-assisted contribution ban, or the Linux kernel disclosure debate. Those were single-project decisions with maintainer essays. Open Source Says No aggregates them so you can see repeat language — “tainted content,” “AI slop,” “generative tools prohibited in external contributions.”
Who is on the November 11 list (and what they say)
The following names appear on the refreshed catalog with verbatim policy excerpts on the site. Star counts come from the catalog’s November 11 scrape.
| Project | Stars (catalog) | Policy gist |
|---|---|---|
| immich-app/immich | 118,789 | Do not open PRs generated with an LLM |
| meilisearch/meilisearch | 59,840 | No generative-AI content in external contributions (code, tests, docs, issues) |
| lysine-dev/okhttp | 47,110 | No LLMs in source or documentation |
| servo/servo | 38,132 | No LLM or probabilistic-tool content in contributions |
| qemu/qemu | 13,843 | Decline contributions believed to include AI-generated content |
| veloren/veloren | 7,656 | LLM output presumed tainted; no commits |
| gentoo (policy) | — | Forbids NLP AI-assisted content in contributions |
Smaller but high-signal repos include stb, Redox OS, MonoGame, Fyrox, pdfcpu, and fast-check (bots and automated accounts not accepted). Several policies grew stricter in October and November 2026 — the same window when Google froze parts of its open-source bug bounty over report volume and when Git may vote on LLM patches.
The November 11 refresh also lists Codeberg projects such as the Zig toolchain mirror (~6.8k stars) with a blunt “No LLMs for patches / pull requests” rule, plus smaller libraries (stb, fast-check, Flask-SocketIO) that illustrate bans spanning languages and contribution types. git.git and Linux are still absent — the catalog tracks explicit published text, not informal maintainer preference threads.
Why maintainers publish bans instead of softer rules
The recurring maintainer argument — visible in COSMIC, QGIS, and QEMU threads explainx.ai has covered — is review asymmetry. A model can emit a patch in seconds; a human still reads diffs, runs CI, and negotiates design fit. When submission cost collapses, backlog grows faster than reviewer hours.
Open Source Says No does not solve that math. It makes the policy visible before you fork or send a coding agent. That is useful for agent operators: if your harness opens unsolicited PRs, repos on this list are high risk for instant closure.
Policies fall on a spectrum we mapped in GHC and peer compiler AI rules:
| Stance | Example from wider OSS |
|---|---|
| Hard ban on generated content | COSMIC, many catalog entries |
| Disclosure plus human accountability | Linux kernel discussion, some enterprises |
| Targeted ban (issues only, docs only) | Varies by template |
| No explicit rule yet | Large swaths of GitHub |
The catalog skews toward the hard ban column because those projects publish unambiguous text the scraper can verify.
Several outline contributors around an open box, illustrating how open-source AI ban policies surface in public catalogs
What developers are saying on Hacker News
The discussion is worth reading alongside the site. Selected themes, attributed to commenters:
- Volume versus ideology: User relyks asked whether bans target unreviewed slop rather than closely human-guided agentic coding — the same distinction American Express workforce automation debates in enterprise, but with unpaid reviewers.
- Misleading headline: lemagedurage argued the name Open Source Says No overstates a small sample dominated by restricted labels in the UI.
- Immich nuance: staminade noted Immich rejects LLM-opened PRs while still shipping agent-related files and Claude-co-authored history — policy is not the same as “no AI anywhere in the repo.”
- Enforcement skepticism: ChrisRR and nsoam said outright bans may encourage hiding tool use or act as legal cover (CYA) without technical detection.
- Counterpoint: Propelloni argued the real filter should be quality, not origin, but admitted maintainers use an AI label as a cheap pre-filter when review time is scarce.
- Author transparency: jlengrand confirmed the site is vibe-coded, early, and intentionally non-conclusive — a research index, not activism.
explainx.ai does not treat HN paraphrases as verified policy. Use them as maintainer sentiment, then click through to each repository’s contributing file.
How this fits next to Codex grants and agent PRs
OpenAI’s Codex for OSS grants push premium coding tools toward maintainers while contribution volume rises from the other direction. A catalog of bans is the receiving side of that squeeze: projects that already cannot review everything agents generate.
If you run an agent that opens pull requests, treat osssaysno.com like a do-not-spam list unless you have a maintainer relationship. Scope work to agreed issues, run tests locally, and prefer repos with explicit allowed agent policies — rare today, but growing in enterprise forks.
For individual developers, the practical question is not “is AI moral for OSS” but whether your target repo’s template will reject the PR before a human reads it. Search the repo for “LLM,” “generative,” or “Copilot” in CONTRIBUTING.md before you spend tokens.
Disclosure templates that survive bans
Projects that allow AI-assisted work with accountability usually ask for three lines in the PR body: what tool was used, what the human verified, and what tests ran. Bans like those in the catalog skip that nuance and jump to “do not submit generated content.”
That is easier to enforce in a template checkbox than a multi-page disclosure form, which is why small and mid-size repos adopt it first while giants argue process. If you are drafting policy for your own library, steal wording only after you read the upstream license — Gentoo’s rule, for example, is a distribution policy, not a generic snippet you can paste into an MIT app.
When your coding agent supports PR descriptions, turn off auto-generated prose for repos on the catalog. Many bans explicitly list descriptions and comments, not just diffs. A human-written three-sentence summary with an agent-authored patch may still fail the letter of the rule even if the code is fine.
Limits of the catalog
- Sample bias: Twenty-two repos is tiny against GitHub’s scale. Absence from the list does not mean AI contributions are welcome.
- Verification lag: Dates on cards show last automated check, not last policy edit upstream.
- Scope confusion: Some projects ban only external PRs, not internal maintainer tool use — read the linked policy.
- Launch bugs: HN reported the suggest button returning 404 at times; use the GitHub issue path the author described.
- Not legal advice: Bans interact with copyright, DCO, and export rules in ways the catalog does not explain — see OpenJDK’s ban for a corporate example.
Maintainer guiding many small contributor shapes with a lantern in an open-source AI policy context
What you should do this week
If you maintain a repo: Decide what you can review. Publish one paragraph in CONTRIBUTING.md, link it from the PR template, and state whether disclosure or certification is required. A ban without review capacity still leaves you drowning — but silent ambiguity leaves you without grounds to close spam.
If you contribute with agents: Check policy text before git push. For catalog-listed repos, assume zero tolerance unless a maintainer waived the rule in an issue.
If you track industry policy: Pair this catalog with the Git LLM vote timeline and Linus Torvalds on kernel review load. Together they show grassroots bans ahead of formal governance on the largest projects.
Related reading
- COSMIC bans AI-generated code · System76’s blunt PR certification rule
- QGIS bans AI-assisted contributions · Geographic information system precedent
- Git project may vote on LLM patches · Debian-style process on the horizon
- Google OSS VRP freeze over AI slop reports · Volume attack on security triage
- What AI slop is · Why quality channels flooded
- GHC and peer compiler AI policies compared · Disclosure versus ban spectrum
- Codex for OSS grants · Tools aimed at the same maintainers
Policy quotes, star counts, and project list accurate as of the November 11, 2026 refresh of Open Source Says No. Individual repositories may change wording without updating the catalog immediately.