Edit: Someone posted they are still going strong. Maybe I just hit an API error.
When the business model rewards higher token usage, almost any good optimization technique should save you money. So I tried one.
Fabric indexes your repo first, finds the files relevant to the task, and gives them to Claude Code before it starts searching.
My tests show up to 62% lower token usage.
Test it here: github.com/hailmarylabs/fabric