Every agent registered — scope set before the first tool call.
Mandates and credentials live in the runtime, not in a spreadsheet your team ignores.
Delegation shrinks at every hop. Revoke once, chain stops — full visibility, no extra workflow.
Fan out to sub-agents without handing them your full authority.
Each child credential is a strict subset of the parent.
Revoke upstream once and the entire chain stops — one operation, full visibility.
Revoke Alex Rivera's credential and both child agents lose access instantly — no orphaned permissions, no manual cleanup.
A research agent with search and read access cannot delegate shell or deploy rights to a sub-agent — because the parent never had them. Scope lives in the credential itself, so validation stays fast without a lookup on every call.
Built on UCAN-style capability tokens — self-contained proof of what an agent may do, validated inline on the hot path.
Revoke the parent and the entire chain terminates.
One operation, no orphaned permissions.
Credentials bind to tool list, model, and config hash.
Change the agent after issuance and the credential invalidates.
Every discovered agent visible in one inventory.
Assign an owner, apply a mandate template, and bring it under governance.
Define tools, data tiers, and time window before the first tool call.
Like scoping a research sprint, a session mandate holds that scope — so agents move fast inside clear guardrails.
Every session gets allowed tools, blocked tools, data tiers, and a TTL. The agent operates inside its mandate — enforcement is structural, not a prompt you hope the model follows.
Mandates evaluate on the hot path before execution. Allows and denies share the same audit record — readable by security and audit without decoding tool codes.
| Field | Scope |
|---|---|
| allowed_tools | search, read_pdf |
| blocked_tools | shell, deploy |
| data_tiers | public, internal_research |
| ttl | 2 hours |
| search("systematic review corpus") | ALLOW |
| shell("install package") | DENY — out of mandate |
Connect your IdP and see governed and ungoverned workloads in one place.
Assign an owner, apply a mandate template, and close the gap before production traffic does.
Every agent discovered across your environment — one view, one audit trail.
Registered with owner, mandate template, and scoped credential.
Discovered agents awaiting owner assignment and mandate template.
Verify tokens from Okta, Azure AD, Google Workspace.
Map standard claims to agent permissions.
Auto-provision and deprovision agent identities from your IdP.
Import dozens of agents from a manifest.
Each gets identity, credential, and owner.
Identity and memory share one fabric — an agent's clearance tiers determine what it can retrieve. Same control plane, two surfaces.
Identity sets who may act. See how inline enforcement and governed memory complete the picture.