01 / The premise
How can an AI coding workflow remember more while reading less?
A local-first context engine for searching source, command output, and research without flooding an AI workflow’s context window.
02 / What took shape
- Token-budgeted result windows
- Persistent named research stashes
- Local-first code intelligence
03 / Looking back
The useful residue.
scrt treats context as a scarce design material. The key shift was from maximizing retrieval to shaping it: the right evidence, at the right resolution, with a durable path back to the source.