Hodor

1 review

AI agents today act with their user’s full rights, and no one defines what the agent itself may do. Hodor (Permission and Policy Engine for AI) fixes that. It sits between agents and business systems as an MCP gateway: Identity: each agent has its own identity, linked to its user, owner and purpose, and synced with the company’s IdP. Least agency: permissions are scoped per tool and per task, within the user’s rights and company policy. Control: every action is checked before it executes, and there is a single kill switch. Audit: every call is logged with who, what and why, and streamed to the SIEM. Data protection: PII is tokenized before it reaches the model. Coverage: 250+ connectors, MCP and OpenAPI, and any model or harness.