Build
An agent is an asset, not a prompt.
Every change produces a new version with a number, an author, a date and a diff. The running version is recorded on every action it takes, so an action from March can be explained by the rules that were in force in March rather than the ones in force now.
Every action records its version
The audit trail stores which version acted, not just which agent. Without that, explaining a six-month-old decision means guessing at configuration that has since changed.
Rollback is one action
If v1.5 behaves worse than v1.4, return to v1.4. The permissions, boundaries and approval rules all return with it, because rolling back the prompt and leaving the permissions widened would be the worst of both.
Changes are described, not just numbered
"Expanded the reminder window from 14 to 21 days. Added approval over €5,000." A version history of bare numbers tells you when something changed and nothing about what.
Widening permissions is a version, and a review
Adding a capability is not a quiet edit. It creates a version, and it returns the agent to the review screen for whoever owns it.
Agent Studio
The rest of this area.
Turn a described job into a configured agent.
Describe the job
Write the job in plain words. No canvas, no nodes.
What gets built
The five things a sentence becomes, all visible before deploy.
Guided creation
Five questions, in the order they actually matter.
Review before deploy
The screen that cannot be skipped.
Ownership
Every agent has a human owner. No orphans.
Give AI a job.
Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.
No code required · Full SDKs and API · Stop any agent instantly · [email protected]