Permissions infrastructure for AI agents
Give AI agents permissions. Not unlimited access.
Define what agents can do across your repositories, infrastructure, and environments. Require approval for sensitive actions, deny forbidden ones, and understand every decision.
See how Gated worksAction-level permissions
One agent. Different permissions.
The same agent can receive a different decision for each action, resource, and environment.
Illustrative policy decision · sample data
Choose a request
release-agentdeployapi-serviceproductionMatched policy
Production deployments require a release owner
Decision
Approval required
Reason
Production changes need a specific human authorization.
Enforcement consequence
The action waits at the gate for a release owner.
The problem
Tool access isn’t permission for every action.
A repository credential, deployment tool, or database connection often permits more than the agent should do. Gated is designed to make the intended boundary explicit.
Repository access
May include
Read an issue
But not automatically
Delete a protected branch
Deployment access
May include
Deploy to staging
But not automatically
Promote to production
Database access
May include
Inspect schema
But not automatically
Delete production data
Platform model
Define the rules. Make the decisions clear.
Gated is intended to evaluate who is asking, what they want to do, where they want to do it, and which policy applies—before a trusted integration enforces the result.
release-agentdeploy api-serviceRequire approvaldeployment integrationPolicy
Rules determine allow, deny, or require approval from request context.
Approval
A specific human authorization for a specific sensitive action.
Decision record
The request context, matched policy, outcome, and reason.
Integration ecosystem
Find your connection to Gated.
Explore planned integrations and how they fit. Initial integrations are planned; these routes describe intended architecture, not active connections.
Evaluates intended request context and policy
Gated decides
Services are grouped by Agent / client, Model / runtime, Action / resource. All integrations are planned.
Planned · Roadmap candidate, not yet implemented.
Private preview · Functioning with limited access.
Available · Tested with a published setup guide.
Developer experience
Put permissions where actions happen.
A conceptual request carries identity, action, resource, and environment. The response explains the policy result and reason; a separate trusted path enforces it.
request
identity release-agent
action deploy
resource api-service
environment production
decision
result REQUIRE_APPROVAL
reason Production release owner requiredThis is conceptual sample data—not a published API or SDK.
Security approach
Clear boundaries. Explicit control.
The intended model separates a policy decision from execution. Control depends on a trusted, scoped path that can enforce the result.
Scoped authority
Limit the authority available to each controlled path.
Decision clarity
Record what was requested, which rule matched, and why.
Coverage limits
Actions outside a trusted enforcement path remain outside Gated’s control.
Early access
Give your agents room to work. Keep control of the actions.
Join the waitlist to follow Gated as the permissions model and first integrations are developed.
By submitting, you acknowledge the privacy notice.