Start with your own logs.
Measure the problem on data you already have, prove the fix with one team, then roll out.
Assessment
We measure how often your assistants break your stated rules, on logs you already have. Runs in your environment; aggregates only; fixed end date. You keep the finding whether or not you buy.
Pilot
One team, 60 days, with success criteria agreed up front.
Rollout
writ for every developer; policy from Canon, enforced on every seat.
Take it with you.
Share them with your team. Technical editions are available under NDA.
writ
Every developer, a team: decisions on every prompt, the record, approvals and enforcement.
PDF · 9 pages
DownloadCanon
One source of truth for every AI agent: policy published to every seat, with fleet evidence.
PDF · 4 pages
Downloadwrit + Canon
The full picture, from one developer’s workstation to the whole organisation.
PDF · 13 pages
DownloadQuestions we hear.
Does writ make the AI’s output correct?
No. It makes the agent’s conduct enforceable, provable and auditable: it works from your decisions, and you can see what it did and what was refused.
Is writ a coding agent?
No. It governs the assistants you already use: Claude Code, Devin and GitHub Copilot.
Is it only for large organisations?
No. writ works for a single developer, a team or a department. Canon adds the organisation level.
Will this slow my developers down?
Refusals carry your approved route, so the agent keeps working. Only weakening a test or running a new tool waits for a person.
How is this different from opening more AI windows?
The same decisions in every window, one record, a shared test platform, a reviewer that isn’t the author, and evidence.
What happens when an agent’s context fills up?
It never reaches compaction. It writes its open work to the record, writ clears the session, and the new one is rehydrated from the record.
Does any code leave our network?
Not to Verilex. writ and Canon run on your machines. The assistants writ governs keep their own model access under your agreements with those vendors.
Can we write our own rules?
Yes. You author both the rule and the approved alternative. writ ships examples, not a mandated policy.
What if writ can’t check something?
It’s reported as unmeasured, never as compliant.
What is the patent status?
Patent pending: a provisional patent application is on file.
Book a demo.
See a team at work, and a refusal, on a workflow like yours.