Agent tooling
ScaleDown's hosted family handles four narrow jobs: assigning labels, copying source passages, summarizing documents and cutting down an agent's context.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
These are competing models, not a Jev integration. Classify returns label probabilities; Extract returns source spans and their positions. The cookbooks show connections to agent workflows. Speed and price comparisons come from ScaleDown, and the hosted family has no published weights.
Pick the cookbook for your task: assigning labels, finding exact passages or shortening documents. When extracting text, keep the original document and the returned positions. That lets your application check that the words really came from the source rather than accepting a plausible new sentence.