Jev in production › SQL and data pipelines

jevgrep

jevgrep asks Jev one yes/no question per piped log or text line and prints the lines whose yes probability clears a threshold, fast enough for tail -f. On its hand-labelled benchmark it scored a mean F1 of 0.90 against 0.87 for Claude Haiku 4.5 (author).

Open on GitHub ↗

mean F1 of 0.90measured against a baseline, as published by the source
Use
SQL and data pipelines
Industry
IT operations
Form
Open-source tool
Stage
In production
Listed
2026-09-22
Found via
discord
Repository
allebee/jevgrep
Stars
2
Forks
0
Last push
2026-09-21
Language
Python
License
MIT
jevgrep screenshot
docs/demo.gif in the allebee/jevgrep README, MIT; shown from GitHub.

The README opens with

grep by meaning. Pipe in any text, ask a yes/no question in plain English, and get back only the lines where the answer is yes.

- Filters by meaning, not keywords. On the benchmark log, it kept the 503 access-log line, the Traceback and the OOMKilled event, none of which contain "error". It dropped errorrate=0.00, 0 failed payments and error handler registered. - Fast enough for a pipe. A match shows up about 0.7 s after the line is written, so it works behind tail -f, kubectl logs -f or journalctl -f. - Cheap. About $0.004 per 1,000 lines. On the benchmark that is 46× cheaper than Claude Haiku 4.5 and 170× cheaper than Claude Sonnet 5, with a mean F1 of 0.90 against their 0.87 and 0.95. - Behaves like grep. It supports -v, -n, --json and pipes,...

Badge

For the project's own README, linking back here:

Listed in Jev in production

Also used for sql and data pipelines