What it is
Jev-native explore loop for open goals.
What it does: The primary AI defines capabilities and completion checks. Jev picks the next admitted tool call from observations and history. Code executes, updates state, and an independent verifier accepts or rejects the outcome.
Why it belongs in the showcase: Separates route discovery (Jev) from tool authorship and verification, with published triage timing/cost evidence.
Caveats: Community project; install from Git, not the unrelated PyPI name.
Evidence
Public MIT repo (1 GitHub star as of intake), docs site, and reports/ evidence pack dated 2026-10-12. README cites a 1,000-alert triage evaluation where a Claude primary chose the records route in 10 of 11 runs (all independently verified), median about $0.27 and 65 s versus about $0.95 and 147 s when reading every alert. Includes browser demo replay and offline contract checks for evidence progress and repeated-action blocking. PyPI name note: install from this Git URL; the PyPI name jev-harness belongs to an unrelated project.
Caveats
- Independent Apixly community project; not a TypeSafe product
- Author-reported eval numbers; calibrate on your own tasks
- Open-goal loops still need host-supplied verifiers and permission bounds
Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.