Authorization
infrastructure
for AI agents.

Keydris lets you control what AI agents can do before they do it. You define a policy that says what an agent is allowed to do. When the agent attempts a governed action, that action is checked against the policy before it runs, and the result is enforced: ALLOW, APPROVAL REQUIRED, or REJECT. You can let an agent read a repository, require human approval before it merges a pull request, or revoke its authority so the next action it attempts fails verification. Revoking does not stop the agent process or undo work it already finished. Keydris makes the authorization decision. The action itself runs in the system that owns it.

No. Keydris is the authorization layer for AI agents. It is not an identity provider, agent runtime, monitoring product, or universal traffic proxy.

You define and change the policy. Keydris issues the agent an authority bound to that policy. Updating or reassigning the policy does not require redoing the agent’s CLI setup.

You can build pieces of the model yourself. Keydris is for teams that want the policy, authority issuance, per-action verification, revocation, administration, and decision evidence to operate as one governed system across supported integrations. The public docs and CLI let you evaluate that tradeoff directly.

OAuth, IAM, service accounts, and workload identity establish who or what has access. Keydris works alongside them and adds a per-action authority check: whether this specific governed action is inside the policy assigned to this agent right now.

MCP authorization and existing identity systems remain part of the foundation. Keydris adds a governed authority lifecycle across the agent, the policy you author, the authority it presents, per-action verification, revocation, and decision evidence.

The receiving boundary evaluates the governed action by requesting verification from the Keydris platform. That boundary can run in your environment, but it does not evaluate the policy by itself.

Keydris does not claim to be a universal proxy for your application traffic. On supported self-hosted paths, payload traffic stays inside your environment and authorization information crosses to Keydris. On third-party systems a Keydris integration may participate in the data path. Keydris does not claim it is never in any data path.

The current platform fails closed: an action whose authority decision cannot be obtained does not proceed. That platform-unavailable behavior is currently fixed, not configurable. It is separate from a policy's own default decision, which you author (for example, reject unmatched actions). Keydris does not claim governed work continues uninterrupted through a control-plane outage.

Not yet. The current public material makes no latency claim. Evaluate the supported architecture and integration paths in the documentation rather than relying on an unsourced performance adjective.

The authority is revoked, not the agent process. The agent may continue running and attempting actions, but revoked authority fails subsequent verification.

A decision record preserves the evaluated action, agent, scope, policy version, checks, outcome, and timestamp for review and export. It is evidence of what the system checked and decided, not a guarantee of universal safety or compliance, and not a copy of the underlying request payload.