IDENTITY
Your IdP stays authoritative
OIDC, SAML, and workload identity. People, services, and agents resolve to identities you already manage — no parallel user store, no second place to offboard someone.
Product
One governed deployment gives every team the same six surfaces: registration, policy, credentials, routing, sessions, and audit. This page walks the surfaces operators run and reviewers sign off on — shown with illustrative data.
The surfaces
Three screens carry most of the governance work: what's allowed to exist, what a policy change would do before it ships, and what actually happened. All data below is illustrative.
The registry is the only front door. Owner, risk, environment, and approval — before anything is visible to an agent.
Simulate a policy change against real principals and tools before it ships — the deny above never reached production.
Metadata only: who, what, when, why, which policy version. Export the whole trail to your SIEM.
The governed path
Every surface above feeds one runtime path. A governed call clears all six stages; a denied call fails closed at the first gate that says no — with a reason, before the upstream is ever touched.
Your IdP stays authoritative — people, services, and workloads all resolve to owned, scoped identities.
Policy evaluates user, agent, surface, and environment. Default deny; deny overrides allow.
The tool list is computed per caller. Unauthorized tools are hidden, not refused later.
The right credential resolved server-side. The agent holds an opaque reference — nothing to leak.
Only registered backends are reachable. Host allowlists and schema checks fail before upstream calls.
A metadata-only record with the decision, the reason, and the policy version — on every call.
> want the full drawing? See the governed-path schematic →
Operator console
Everything the console does, the CLI and admin API do too — so registration, policy simulation, and rollout live in version control and CI, not in someone's browser tabs.
OpenAPI and JSON Schema contracts define every control-plane action. If the UI can do it, your pipeline can do it.
Manifest validation, registration, and policy simulation run in CI — a bad manifest fails the build, not the incident review.
Machine-readable reason codes with suggested fixes. An agent that gets denied knows why — and what approval to request.
Built to be trusted
The platform choices are the security story: components your platform team already runs and knows how to back up, patch, and monitor. Releases ship signed, with an SBOM you can scan before anything reaches your cluster — and self-hosted deployments make no calls back to us.
> honest note: formal certification (SOC 2 and similar) is on the roadmap — we'd rather tell you that than imply otherwise. Full security posture →
Fits your stack
Identity, secrets, and evidence stay where they already live — the gateway consumes them and adds the agent-to-tool boundary they don't cover.
OIDC, SAML, and workload identity. People, services, and agents resolve to identities you already manage — no parallel user store, no second place to offboard someone.
Credentials resolve at call time against your Vault-compatible secret store. We don't replace your secret manager — and agents never see the material, only opaque references.
OpenTelemetry for traces and metrics, SIEM export for audit records. Evidence flows into the dashboards and alerting your security team already watches — no new pane of glass.
Next step
Bring your IdP, your secret store, and one boundary you'd govern first — we'll walk the whole path on a working deployment.