Hey Product Hunt 👋 We're Corbenic, and we build with Claude Code and Copilot all day. Watching the token bill climb, we noticed something: agentic tools resend the entire conversation every turn, including pages of the model's own old "thinking" it never needs again, plus context repeated byte-for-byte. You pay for all of it, every single turn. So we built Aneirin: a small local proxy that strips that redundant, re-billed context before the request ever leaves your machine. It's lossless. The model receives everything it needs, your answers don't change, you just stop paying twice for the same tokens. On our own machines it's removed 650M+ tokens so far, and a typical optimized request comes out about 17% smaller. Two things mattered most to us. First, it runs 100% locally with no phone-home, so your prompts and code never leave your computer (no telemetry, and the licence even checks offline). Second, it's genuinely free to use: the full optimizer runs free up to 5 million optimized tokens a day, no account needed. Pro just removes the daily cap. We'd love your feedback, especially edge cases where you'd want it to behave differently. Happy to answer anything about how the lossless part actually works.