What it does
A local-first layer that sits between your coding agent and its context. It selects and compresses what is sent to the model and keeps a receipt of what was left out, so the original can be recovered. It can also route work to cheaper models and run local hallucination checks.
Use cases
- 01Trim the context a coding agent sends on each request
- 02Recover the exact original text behind a compressed selection
- 03Route simpler tasks to a cheaper model