Every team has rules. Agents should know them.
Coding agents now write a large share of the code. They're fast and capable, and they only follow what they're told. Meanwhile a team's real rules live in review comments, a wiki and a dozen copies of AGENTS.md. Groundrule gives every team one rulebook that every agent reads and every change is checked against.
About two weeks, from a review comment to a check in CI.
How one rule travels.
- PR #470A review comment
“We don't call Stripe directly. Use the gateway.” Said for the fourth time.
- todayA proposal
/groundrule rule turns it into an inbox item, citing the comment. The Payments owners see it first.
- ACME-004A standard
Accepted, drafted with a check that allows only src/payments/gateway.ts. Published at Teach.
- next taskEvery agent knows it
sync writes it into AGENTS.md, CLAUDE.md and Cursor rules in every repository.
- day 14Every change is checked
Clean everywhere for two weeks, it moves to Enforce. CI stops the next direct Stripe call.
Review doesn't scale to agents.
The bottleneck moved from writing code to agreeing on it.
For years, a team's standards travelled the slow way: a senior engineer explained them in review, one pull request at a time, and new people absorbed them over months. It mostly worked, because people write code slowly and remember what they're told.
Coding agents changed both halves. They write in minutes what took a day, and they start every task knowing nothing about your team. They'll call Stripe directly, log a session cookie or store money as a float, because nothing told them not to. Reviewers become the only place the rules live, and review doesn't scale to agents.
The fix isn't a smarter reviewer. It's writing the rules down once, in a form people and agents can both follow, putting them where agents look, and checking the result automatically. That's slow to do by hand: the rules are scattered, nobody wants to write them from scratch, and a new check that turns CI red on day one is a check people learn to ignore.
So Groundrule starts from maintained packs, finds the rules you already have, lets the right people review them, and rolls each one out in stages, on evidence from your own code.
If we ever have to break one, we'll tell you first and give you time to leave.
What stays true, whatever we build next.
Your rules stay yours
Standards are YAML in an open, documented format. Export them, keep them in a repository, use the open-source CLI without us.
A person decides
AI reads, drafts and suggests. It never adds, changes or removes a rule on its own, and everything it writes says AI.
Your code stays where it is
Checks run on your machines. We receive results, never your source, and you can turn off even short snippets.
Nobody blocked by surprise
A rule fails a build only after someone moves it to Enforce, and you see what it would catch before you do.
The status page tells the truth
What works, what's in progress and what isn't built yet, kept current by hand. If it's wrong, that's a bug.
Least access, everywhere
Read-only connections, scoped tokens, and a database that keeps workspaces apart by itself.
Tools in this space promise a lot. These are the claims we avoid, and the ones the product backs up.
What we won't say, and what we will.
- “Your code is secure” or “compliant”. Checks show where rules pass and fail; compliance is a judgment for you and your auditors.
- “AI enforces your standards.” AI proposes. Deterministic checks and your team enforce.
- “Zero false positives.” Every standard carries a noise rating and lists where it misfires.
- “Set it and forget it.” Rules need owners. Groundrule gives them a place to work.
- Every agent in every connected repository reads the same rules, at the stages you chose.
- Every finding names its rule, the file and line, and how to fix it.
- Every AI result cites its source and states its confidence.
- Before a rule can fail a build, you can see exactly what it would catch.
Write to us; a person answers.
Built by the people who answer your email.
There is no support desk in between: the people who write the code read every message. The open-source kernel is developed in public on GitHub.
| General | hello@groundrule.dev |
| Support and security | support@groundrule.dev |
| Open source | github.com/Metricall-AI-Lab/groundrule-oss |
| Documentation | docs.groundrule.dev |
Give every agent your team's rules.
Create a workspace, adopt the packs that fit your stack, and connect one repository. In half an hour your coding agents follow your standards, and you can see what a check would catch before it blocks anyone.