Four pillars. One control plane.
Identity, policy, audit and a kill switch — the same primitives, whichever agent, whichever vendor.
No anonymous agents.
Every agent gets a cryptographically signed identity bound to an owner, a purpose and a vendor. Credentials are short-lived and scoped per task — never a raw API key sitting in an environment variable.
- Signed identity per agent, tied to a human owner
- Short-lived, per-task scoped credentials
- Inherits org structure from Entra and Okta
| Agent | Vendor | Owner | Credential |
|---|---|---|---|
invoice-reconciler | OpenAI | MO | expires in 04:12 |
tier1-support-triage | Anthropic | RK | expires in 11:47 |
sev2-oncall-responder | Microsoft | unassigned | blocked · no owner |
pipeline-janitor | In-house | DP | expires in 00:58 |
Say the rule out loud. FleetWarrant enforces it.
Write the policy the way you'd say it in a risk meeting. FleetWarrant compiles it to a structured, enforceable rule — versioned in git, reviewed like code, evaluated on every single action.
Revoke everything. One button.
One button revokes an agent's credentials across every connected system, in seconds. Circuit breakers trip automatically on policy violations and anomalous behaviour, and page the owner before you ever open Slack.
Space or Enter also works, held down.
Every tool call, forever immutable.
Searchable in seconds. Replayable step by step.
Live audit log of agent actions. Each row shows a timestamp, the acting agent, the action taken, its target, latency and policy verdict. New events stream in continuously; this description is not repeated for every event.