Lobster Relay

Lobster Relay is a lightweight orchestration layer built around Veea’s Lobster Trap to improve how AI agents handle policy enforcement failures. The project focuses on one specific problem: most agent systems treat blocked requests as dead ends. When an action is denied, agents typically receive little context about what failed, why it failed, or how to continue safely. Lobster Relay adds structured recovery on top of Lobster Trap’s enforcement pipeline. Safe requests pass through normally. When Lobster Trap intercepts a request, Lobster Relay generates a concise, explainable response that helps the agent understand the failure and retry in a compliant way. This creates a more usable interaction pattern for autonomous systems: attempt → fail → understand → adapt → retry The system also includes: OpenAI-compatible relay APIs LiteLLM-based provider routing Runtime interaction logging Real-time audit visibility Structured explainability workflows Lobster Relay is designed to complement Lobster Trap by improving recoverability and usability for governed AI agent workflows.