AI-built websites are lawsuits waiting to happen.
Bellbriar teaches your agents the rules and proves they followed them. Attorney-reviewed knowledge, loaded on every build.
- status
- signed · reviewed
- diff
- +14 -3 since last release
- graders
- axe-core + model judge
- eval
- 41% 99% pass
- backing
- guidance reviewed by counsel
The lawsuits are already happening. AI just scales both sides of them.
Compliance widgets cover up mistakes, like covering a hole with a picture. The code underneath is untouched — which is what gets read in court. We fix it upstream, measure the changes, and give you the receipts.
You can generate AI skills at scale — but you can't generate trust.
Compliance content is everywhere, and mostly free. But it doesn't tell you where it came from, when it was checked, or whether it actually worked.
Where it came from
Every Bellbriar pack is reviewed by an attorney against primary sources, then signed and dated. It's expert human judgment, not AI-generated text about AI-generated text.
When it was last checked
Compliance content goes stale the day you download it. Our pipeline watches the rules and ships a reviewed release when they change — an update instead of a surprise.
Whether it worked
Every Bellbriar pack ships with an eval harness that measures how often agents pass with it and without. We publish our own numbers on it — see the benchmark.
npm for compliance knowledge.
Pin the version. The eval runs in your CI as a gate on agent output. When the rules move, it arrives as a dependency bump — reviewed, signed, and diffed against the last release.
Semver releases, signed artifacts, changelogs, version pinning.
Drop it in your repo, or call it over MCP and always get the current release.
# load the reviewed pack your agents build against
bellbriar add wcag@2.2.1
# gate every build on the eval harness
bellbriar eval ./dist --report report.json
status: signed diff: +14 -3
graders: axe-core + model judgeIllustrative surface. Interface details may differ with version used.
From free audit to continuous cover, in four steps.
Send ten links. We run them through the harness and send back every criterion that fails.
Your agents read the pack on every build. Compliant patterns become the default output.
The eval runs in CI. Output that regresses below the bar doesn't ship.
When rules change, you get a reviewed release and a re-scored portfolio.
Compliance knowledge, versioned like software.
Web accessibility — ADA / WCAG
Where the rules are testable and the filings are heaviest. If your agents build public-facing pages, this is the exposure you have today.
NYC real-estate compliance
Rent stabilization, the FARE Act, brokerage licensing — rules that change often and are expensive to get wrong.
Upcoming Packs: HIPAA · PCI · ad-claims law · building codes
Find out what your pages are carrying.
Send up to ten links. We'll audit them and send back every criterion that fails — the specific list, not a grade.