JevMade Sign in
← Back to experiments

Agent tooling

jevgrep

A grep that matches meaning instead of words, written for live log files.

Bookmark: jevgrep Keep this in your collection.
Leave a noteWhat would you try with this? : jevgrep

Only you can see your notes.

Source screenshot of jevgrep
SOURCE SCREENSHOTFull screenshot ↗

What it does

Give it a question like 'this line reports a real error' and it keeps the lines that qualify, quick enough to sit behind tail -f. The benchmark includes the cases it misses.

How you can use it

First, gather a small sample of your server logs. Write down the exact yes-or-no question you want answered, including any borderline examples you want ignored. A developer can install the tool and connect it using an account access key from OpenRouter or TypeSafe.

Be careful with sensitive records. The tool sends every scanned line across the internet to outside servers for evaluation. Never feed in user passwords or secret files. Sneaky text inside a log can also fool the system, so always double-check matches before triggering automatic fixes.

Maker-reported (not independently measured by JevMade): ~$0.004 per 1,000 lines; mean F1 0.90 against Claude Haiku 4.5's 0.87 on the benchmark log (maker-reported)

Primitives
noul
Platform
Python
Added
Project created

Keep this for later

Sign in to bookmark experiments, guides and videos, and keep notes only you can see.

Continue to sign in

We’ll bring you back to this listing.