Platform · Enterprise controls
Built for agents running in production.
The spec sheet for platform, security and procurement teams.
01
Runtime
- Decision latency
- Designed to sit in the request path of production agents without slowing them down.
- Fail-closed behaviour
- If the checkpoint can't be reached, actions are denied rather than allowed.
- Framework support
- Any agent framework or model provider, through an SDK or an API gateway.
02
Change control
- Versioned policies
- Every change is a new version, reviewed before publishing and reversible in one step.
- Dry-run
- Replay past agent activity through a draft policy before it goes live.
- Environments
- Separate policies for development, staging and production.
03
Access
- Role-based access
- Control who can write, review, publish and approve.
- Single sign-on
- Sign in with your identity provider.
- Planned
04
Data
- Append-only audit log
- Every request and decision, exportable to your security and data tools.
- Minimal collection
- Records the actions and decisions it needs — not copies of your models or training data.
- Regional hosting
- Choose where decision data is stored.
- Planned
Security
Where we stand on security and compliance today, and what we're working towards.
From the blog
Related reading
- EngineeringDeploy AI agents to production: a readiness checklistBefore you deploy AI agents to production, check identity, limits, approvals, testing, monitoring and rollback. A practical readiness checklist.
- EngineeringHow to test AI agent guardrails before productionTest AI agent guardrails by replaying real agent activity through a new rule before it goes live, so limit changes never surprise anyone.
- ComplianceAI agent compliance: the controls and records reviewers ask forAI agent compliance comes down to provable controls — enforced limits, human oversight and complete records. What reviewers ask for and how to be ready.


Planning a production rollout?
Talk to an expert about your architecture, review process and timeline.