Sharing an open-source project I've put ~4.5 months into (disclosure: I'm the maintainer; per the self-advertisement rule I'm keeping the link in the first comment and making this post substantive). It started from two problems I hit daily: AI runs dying on a provider rate limit, and burning thousands of tokens dumping tool/log output into the context window. One endpoint, 237 providers — 90+ of them free. You point any tool or agent at a single OpenAI-compatible endpoint ( localhost:20128/v...