Launched this week
WRUD
See what your AI agents did - and where they overpaid
17 followers
See what your AI agents did - and where they overpaid
17 followers
wrud is an open-source CLI that records what your AI agents did each session - tools, files, models, tokens - to a local SQLite file, and flags where a frontier model did trivial work so you can right-size. Claude Code & Cursor built in, or any agent via the SDK. No cloud. npx @wrud/cli
how does it actually detect when a frontier model is doing something trivial vs legitimately complex work? curious what signals it's looking at under the hood
Ran it on a Claude Code session this morning and the token breakdown was eye opening, I had no idea how much of my spend was just a Haiku call reformatting logs. Solid little tool, love that it's all local.
@sibeltcwa thank you, happy for that!
ran it against a long Cursor session and the SQLite output was surprisingly readable, easy to grep through and see which calls were wasted on simple tasks. local-only and no signup is a nice touch.
Ran wrud against my Claude Code session and the "trivial work" flags made me laugh - apparently I had Haiku summarizing one-liner edits. Local SQLite feels right, nothing leaving my machine. Solid little tool.
Helped me to understand the real worth of my Claude Max 200$ subscription. Thanks for that.
@itsgaldoron happy to hear that!
This is super useful! I automate my daily tasks with AI, so I burn through a lot of credits. But since I’m not staring at my screen all day, I couldn't see which processes were taking up the most time, meaning I was often wasting credits. This looks like the perfect solution to that exact problem. I'm definitely going to try it out!
@ryutaro_imai Waiting to hear your feedback! thanks!