Retrieval heat 3 · Mild★ 50heat incomplete (no X)

jevfetch

A MIT web-reading CLI/agent tool that splits a page into blocks and asks hosted TypeSafe Jev which blocks answer a question, returning verbatim passages with heading paths instead of a paraphrase.

What it is

Quote the page; do not invent a summary.

What it does: Block-splits HTML, scores answer relevance with hosted Jev, and returns kept passages verbatim.

Why it belongs in the showcase: Strong evidence trail for hosted-Jev extraction in agent web reading.

Caveats: Benchmarks are snapshots.

Evidence

Public MIT repo, created 2026-10-08, 50 stars as of 2026-10-30. README compares jevfetch vs Claude Code WebFetch on documented Korean/English and Exa WebCode Highlights runs, links bench/RESULTS.md, and states Jev judges blocks. Distinct from crawl/search wrappers that paraphrase.

Caveats

  • Bench numbers are dated agent runs; re-check before citing as current product claims
  • Needs network fetch of the target URL plus TypeSafe spend
  • Pages that block automated fetch still fail open

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.