Claude Code and GitHub Copilot resend your whole conversation every turn, including pages of the model's own old "thinking" and context repeated word-for-word. You're billed for all of it. Aneirin is a local proxy that strips the redundant, re-billed tokens before the request leaves your machine, losslessly: the model still gets everything it needs, your answers are identical, your bill drops. 100% local, zero telemetry, your prompts never leave. Free up to 5M optimized tokens a day, no account.
