Mohammed Muteeullah — blog https://muteeullah.b0lt.foo/blog/ SRE, homelab, and LLM tooling notes. I benchmarked two agent token-compression tools on real data — the higher ratio was a traphttps://muteeullah.b0lt.foo/blog/rtk-vs-headroom/https://muteeullah.b0lt.foo/blog/rtk-vs-headroom/Tue, 07 Jul 2026 00:00:00 GMTRTK vs headroom on real coding-agent tool output. The biggest per-command compression ratios came from throwing answers away — and stacking both tools made parity worse, not better. A parity-first way to decide what to run.