Three readers, one record.
Security teams see and stop risky tool calls. Builders ship agents with approvals and evidence built in. Reviewers read the trail instead of chasing screenshots.
| Reader | Question answered | First action |
|---|---|---|
| Security teams | Which agents reach which systems, what was stopped and why | Connect one tool; watch the record |
| Agent builders | How to ship with identity, policy, approvals, and receipts built in | Dry-run policy before enforcing |
| Reviewers | What happened, with sourced and exportable proof | Read the trail; mark open questions |
Security teams: which agents reach which systems
See registered and observed agents within the configured coverage, the systems each one can reach, and who approved it — then answer what was stopped and why, straight from the trail.
First action: inventory one app through a connector, then instrument one tool path with an SDK guard or MCP route to evaluate enforcement.
- Morning: open the record and see registered and observed agents within the configured coverage, the systems each one can reach, and who approved it.
- Midday: a risky call waits for approval. A person reviews the request, the matching rule, and the recorded reason.
- Friday: answer what was stopped and why straight from the trail, with a signed receipt behind each decision.
Agent builders: approvals without the plumbing
Give the agent an identity and a narrow scope; inherit policy checks instead of writing them. The gate behind all three readers is on the gateway page.
First action: read how the gate decides on the policy page, then prototype against it — Studio implementation and limits are on the Studio page.
- Design: give the agent an identity and a narrow scope; inherit policy checks instead of writing them.
- Test: dry-run the policy and replay past calls before anything enforces — see what would be denied while it is still cheap.
- Ship: policy and session constraints can hold high-risk calls for review; inspect signed receipts and their verification status.
Reviewers: read the trail, not the chat logs
Ask which agents acted in a window and under whose approval; inspect exported reports and proof bundles. External research cards on this website have separate primary-source citations. Open questions stay labelled as open.
First action: see what we hold ourselves to on the trust page, then check dates on the roadmap. Talk to us: get in touch.
Short answers to common questions are on the FAQ page.