What is Policy Hot Reload?

In production environments, restarting a proxy to update configuration is unacceptable. Active MCP sessions would be interrupted, in-flight tool calls dropped, and agents left without their tool infrastructure until the proxy comes back up. Hot reload eliminates this operational burden — you edit the policy in the dashboard, save it, and the new document applies to the next tool call.

Hot reload also changes how teams work with policies. Without it, policy updates require a deployment — pull request, CI/CD pipeline, restart. With hot reload, a security engineer can respond to an incident by editing the grant's policy in the dashboard and having the change take effect in seconds. This operational agility is critical when an agent is misbehaving and you need to restrict its access immediately.

The mechanism is save-time validation. When a policy is saved in the dashboard, the document is checked against the policy schema before it becomes active. If the new document is invalid (structural error, schema violation), the save is rejected and the gateway continues enforcing the previous valid policy. This ensures a mistake in a policy edit never takes down the enforcement layer.

PolicyLayer policies are JSON documents attached to grants and enforced at the hosted gateway. When a policy is saved in the dashboard — whether edited in the visual policy builder or pasted into the Raw JSON view — the document is validated against the policy schema, and on success it becomes the active policy for that grant. Each tool call is evaluated against exactly one version of the document, so in-flight evaluations are never affected. If validation fails, the save is rejected with details of the issue and the previous policy remains active.