Brex Launches CrabTrap: LLM-as-a-Judge HTTP Proxy for AI Agent Security
Brex has introduced CrabTrap, an HTTP proxy system that uses large language models as judges to secure AI agents in production environments. The tool intercepts every request made by AI agents, evaluates them against defined policies, and makes real-time decisions to allow or block these requests. Setup can be completed within 30 seconds using provided terminal commands. The system provides visibility into its decision-making process, logging whether each action resulted from static rule matches or LLM judgments. Users can edit existing rules or create new ones to test CrabTrap's interception capabilities. Brex positions this development as proactive engineering that addresses gaps in the AI ecosystem rather than waiting for industry solutions to emerge. The company is actively recruiting engineers who share this build-and-ship approach to problem-solving.
Key facts
- CrabTrap is an LLM-as-a-judge HTTP proxy
- It secures AI agents in production
- Intercepts every request AI agents make
- Evaluates requests against policies in real time
- Allows or blocks requests based on evaluation
- Setup takes 30 seconds with terminal commands
- Logs decisions as static rule matches or LLM judgments
- Users can edit and add new rules
Entities
Institutions
- Brex