Brex Open-Sources AI-Powered Proxy for Secure Agent Access

Brex, a fintech company, has made its internal security platform CrabTrap available as an open-source project. The platform is designed to control how autonomous AI agents access APIs, credentials, and online services. This move comes after Brex explored the deployment of agent frameworks like OpenClaw across its operations.

The company found that traditional model guardrails and tool permissions were insufficient for controlling agents with access to sensitive information such as API keys, OAuth tokens, and service accounts. To address this issue, Brex developed CrabTrap, an HTTP proxy that intercepts AI agent network traffic and uses a large language model (LLM) judge to approve or deny requests in real-time.

CrabTrap operates by examining traffic against established policies before deciding whether a request should be approved or blocked. The system combines deterministic rules with the LLM judge’s evaluation of requests outside known traffic patterns. Brex notes that the AI judge typically activates for fewer than 3% of requests after an agent has established predictable behavior.

Users can route their agent traffic through CrabTrap by configuring their HTTP_PROXY and HTTPS_PROXY environment settings, allowing it to work across different programming languages, frameworks, and APIs without requiring individual SDK integrations. This flexibility makes the platform a valuable resource for businesses looking to leverage AI tools in their operations.

Brex also built a policy generator that studies historical network traffic while agents operate in shadow mode. The system samples requests and drafts natural language policies based on how the agents actually behave, providing a comprehensive approach to agent management.

One of the main development challenges was prompt injection, where malicious URLs or request bodies could attempt to manipulate the LLM judge’s decision. Brex addressed this issue by converting requests into structured JSON before passing them to the model, ensuring user-controlled content is escaped rather than inserted as raw text.

The company has reported that CrabTrap has increased internal confidence in deploying autonomous agents across more business operations and helped identify unnecessary tools, requests, and token usage within its agent systems. Brex plans to expand the platform with features such as single sign-on, role-based access controls, permission escalation workflows, and automated policy management.

The open-source project has already attracted significant interest from developers and businesses alike, including OpenAI, Y Combinator CEO Garry Tan, and developer Pete Steinberger. With its ability to provide secure agent access using an LLM as a judge, CrabTrap marks a significant step forward in the development of AI tools for business.