Condense.chat Proxy Cuts AI Coding Agent Bills by 66%

Condense.chat has launched a new proxy system to significantly cut AI coding agent bills. This system sits between an agent and its model, shrinking each request before billing. What's interesting is that this proxy can reduce costs by 66 percent on average. It works by compressing content, especially the repetitive parts of long coding sessions. For example, in one session, the bill dropped over 72 percent, turning a $154 run into $43. The proxy uses two compression models, Helene 1 and Adeline 1, to either strip content or summarize past loops. It preserves essential information like system prompts and final answers. Setup is simple, requiring a single command. The proxy supports Claude Code, Codex, and OpenCode across different operating systems. This matters because it can save developers running large-scale agent workflows a lot of money.