Show HN: I built a smart proxy so your coding agent can run loose (trollbridge.dev)

A developer has introduced a new tool called "trollbridge," designed to enhance the security and management of outbound coding agents. Unlike traditional interception toolkits like mitmproxy, trollbridge acts as a policy-enforced gateway that integrates an optional LLM (large language model) advisor to make real-time decisions on HTTP requests. By leveraging inline allow/deny lists, the majority of traffic is filtered before any AI is consulted, ensuring efficiency and minimizing costs. If a request requires further analysis, it is classified by the LLM advisor, which only operates when explicitly configured to do so. The significance of trollbridge lies in its ability to provide detailed logging and human oversight without compromising performance. Each LLM decision is recorded in an audit log, allowing operators to review model outcomes alongside their confidence levels, ensuring transparency. Furthermore, trollbridge can operate entirely offline, avoiding third-party LLMs by enabling users to deploy their models locally. This combination of policy enforcement, granular request handling, and optional AI involvement positions trollbridge as a sophisticated solution for managing coding agents in a secure, controlled manner.