Skip to content

AI Agent Governance

AI agents can read databases, send emails, call billing APIs, and modify infrastructure. Governing what they are allowed to do - and keeping a record of what they actually did - is what the KeyRunner Control Plane is for.

What you can do here

  • Register agents and define what APIs they are allowed to call
  • Set policies that control which tools each agent can use
  • Require human approval for sensitive or high-risk actions before they run
  • Review every execution in an immutable audit log
  • Validate an agent's configuration before deploying it to production
  • Test whether a specific tool call would be allowed or blocked, without running it for real

How it works

When an agent attempts a tool call, KeyRunner checks it against the agent's policies before anything executes. The check takes milliseconds and produces one of three outcomes:

Allowed - the call meets all policy requirements and runs immediately.

Approval required - the call is held (or flagged for review) until a designated approver signs off.

Blocked - the call does not meet policy requirements and is rejected. The agent receives a clear reason.

Credentials are never exposed to the agent. KeyRunner retrieves them server-side and executes the API call on the agent's behalf.

Key concepts

Agent - an AI process you have registered with KeyRunner. An agent needs at least one policy before it can call any tools.

Tool - a specific API action agents can request (for example, github_create_issue or stripe_charge_customer). Tools must be registered and approved before agents can use them.

Policy - a named set of tools. Attaching a policy to an agent grants that agent access to all tools in the policy. You can optionally add approvers to a policy, which routes every covered tool call through human review.

Approval request - a queued tool call waiting for a human decision. Approvers are notified via Slack and can approve or reject with a reason.

Pages in this section

PageWhat it's for
AgentsRegister agents and manage their identity, policies, and status
ToolsReview and approve API actions agents can request
PoliciesDefine which tools agents can use and who must approve them
ApprovalsReview and act on tool calls waiting for human sign-off
Audit LogsSee the full history of every tool call across all agents
Agent GraphVisualize your agent fleet and each agent's governance chain
PreflightCheck that an agent is correctly set up before going to production
Policy PlaygroundTest whether a specific tool call would be allowed, without running it
Governance ConceptsReference for identity, lifecycle shapes, and enforcement rules
Connecting Your AgentConnect your agent to KeyRunner using the SDK

Released under the MIT License.