Play the video in the post belowPost on X
- Views
- 661,032
- Likes
- 3,939
- Reposts
- 268
Read from X on . Counts change daily.
newsjack is a set of open-source agent skills for PR work: monitoring news for stories a company can jump on, checking whether a story is still fresh, finding journalists and drafting pitches. It installs into agents such as Claude Code and Codex. Jev comes in through a coarse relevance filter, where newsjack coarse-filter --engine jev turns the public relevance rubric into typed questions.
The video Elvis posted comes from News Desk Dealer, a demo in the repo that has Jev and Claude Opus 5 judge the same Google News headlines side by side. One question set tags each headline with a desk, a story type and six score questions on a 0 to 4 scale (magnitude, velocity, novelty, window, heat, risk), which feed a newsworthiness score out of 10. A second set scores each story against 15 example companies on standing and journalist shape. When Jev finishes, the Opus run stops.
In the posted run, Jev read 384 headlines from one morning and told 15 brands which stories to act on in 24.9 seconds for $0.19. Opus 5, on the same feed at the same time, got through 4 of 384 for $0.77, which Elvis works out to about 390 times cheaper per headline. The demo’s README reports about 190 milliseconds per call for the first question set and 320 for the second. The demo runs on a mock engine by default; the live Jev and Opus engine needs TypeSafe and OpenRouter keys.
