Tejas Chopra
Tejas Chopra
The concern about prefix cache invalidation is totally valid and worth addressing directly — so let me explain exactly what happens. The short version: Headroom does NOT touch your cached...
Note - Headroom shows the prefix cache hits and misses - And this is from the LiteLLM response itself. The response captures how much prefix cache hit we have seen...
@TyceHerrman Headroom forwards all Claude Code headers (auth, telemetry, session IDs) verbatim to Anthropic — we don't replace the harness, we compress inline. Only the latest unfrozen message is touched;...
Hi - thanks for this change - can we resolve some conflicts.
Split out of #818. Audit-reads (the offline simulation that sizes this mechanism) stays in #818; this PR carries only the live transform + wiring + tests.
Thank you - great fix!
Fantastic @orty - so in essence - this works on your Windows - with your Enterprise Github Copilot License? Can we confirm that? Also - do you see token savings?
Hi - can you help fix the conflicts for this? Thanks!
I believe this is fixed with latest merge on main?
This will need a deep review - any chance this can be broken into multiple PRs? Also - i see a lot of fixtures/*.json files - can you help me...