A YC-backed team has released Context Gateway, an open-source proxy that sits between AI coding agents and large language models to compress tool outputs before they enter the context window. The tool addresses a critical problem: agents like Claude Code and OpenClaw often dump thousands of tokens into conversations, most of which are noise that degrades quality. Long-context benchmarks show steep accuracy drops as context grows.

Context Gateway uses small language models to analyze model internals and train classifiers that detect which parts of the context carry the most signal. When a tool returns output, the system compresses it based on the intent of the tool call—so if an agent calls grep looking for error handling patterns, the SLM keeps relevant matches and strips the rest. The proxy also provides spending caps, a dashboard for tracking sessions, and Slack notifications when agents are waiting.

Developers can install Context Gateway with a simple curl command and configure it through an interactive TUI wizard. The tool supports Claude Code, Cursor, OpenClaw, and custom agent configurations. By compressing history in the background and only showing tools relevant to the current step, Context Gateway eliminates the waiting that typically occurs when conversations hit context limits.