Jev
Projects and libraries using Jev for routing, verification and typed decisions.
Repository owner imageWinnow
Reduce irrelevant tool output and select task-relevant memory for coding agents.
CategoryClassification & routing | UseCoding-agent context selection | SourceGitHub |
Ownerghalebdweikat | Original projecthttps://github.com/ghalebdweikat/winnow | |
Research evidence
TypeSafeJudge in sidecar/src/winnow/judge.py calls system_one. hooks.py asks block-relevance questions, caches the original tool output before replacing low-relevance sections, and separately asks memory-file Nouls to select prompt context. policy.py keeps missing or uncertain block scores and preserves all output when the error-present signal is high.
One independently documented plugin/sidecar, with filtering and memory selection as related features. The optional generative System One adapter and summarizer are not official Jev execution.
Judge exceptions pass tool output through; this is a relevance aid, not a security boundary. Cache and logs can retain original sensitive output. Benchmarks and recall preservation were not reproduced; no hooks were installed or executed.
documentation source 1 (retrieved 2026-09-18)code source 2 (retrieved 2026-09-18)code source 3 (retrieved 2026-09-18)code source 4 (retrieved 2026-09-18)code source 5 (retrieved 2026-09-18)