nua (fka Rudel)
Claude & Codex session analytics that improve agent skills.
Start new thread
Marc Höffl

Rudel - Claude Code & Codex session analytics for dev teams

by
Laserfocus
•
We connected 1,573 real Claude Code sessions and found that skills fire in only 4% of sessions, 26% get abandoned within 60 seconds, and error patterns in the first 2 minutes predict failure. So we turned it into Rudel and release it to the world. Rudel gives your dev team a dashboard for token usage, session patterns, session errors, and quality signals across every team member. Works also for Codex. It's free, open source and self-hostable.

Add a comment

Replies

Best
Marc Höffl
Maker
📌
Hola hunters! 👋 We built Rudel because we were embarrassed. We spend $1.4k per month on Claude Code and are using it every day. But we couldn't answer the most basic questions: Which sessions actually moved the needle? Why did half of them just... die? Were we getting smarter with context management, or just lighting tokens on fire? What skills are missing? Are we using plan mode too much? So we instrumented ourselves. After like 1,573 sessions, 15M+ tokens, 270K+ interactions we found out: - Skills triggered in only 4% of sessions even if configured. Claude (and Codex) just didn't use them... - A quarter of sessions get abandoned , most within the first 60 seconds - Error cascades in the first 2 minutes are the clearest predictor of abandonment We launched on HN some weeks ago and got 144 points, 72+ comments. Turns out we were not alone, so we're launching this on Product Hunt now too! Three commands to get started: `npm install -g rudel` → `rudel login` → `rudel enable` Fully open source. Self-hostable if you want to keep session data on your own infra. We'd love your honest feedback. It is super early. Embarrasingly early Now dig in and start rudeling.
George Roberts

@marc_hoffl super cool idea! It just looks directly in the claude code/codex sessions? It doesn't need API access or anything?

PS. Clicking the buttons on your website was very satisfying

evren
Maker

@george_roberts2 nope, no need for API. you run the commands within the repo you want the sessions uploaded and they're all within rudel :)

evren
Maker

@george_roberts2 ps. thx haha was fun building that. thought about doing a leaderboard of ppl who clicked most on the buttons but then i only had 2 hrs until the launch deadline lol

evren
Maker

building that is so much fun

Christophe Dupont

That 26% abandoned in 60 seconds stat hit me. I've definitely rage-quit sessions when Claude starts confidently rewriting files I didn't ask it to touch. Curious about the error pattern detection though. Does it catch the classic 'Claude apologizes and then makes the exact same mistake again' loop? Because that's where most of my tokens go to die. Gonna self-host this, $1.4k/month deserves a receipt.

Lucas García de Viedma Pérez
Maker

@thenomadcode hey hey! At the moment error detection focuses more at the cases where the agent wants to invoke any kind of tool or feature of the harness without success. We do have logic that categorizes (tags) sessions based on what happened in them, so doing so for errors should also be pretty straight forward. Would be amazing if you wanted to contrubute to that with a quick pr!

Alex Smith

Does this support subscriptions or only API usage?

Álvaro
Maker

@youmeapps It supports subscriptions too, and counts the tokens the same way as if you were paying for the API key. It's nice to understand how far your subscription can get you.

evren
Maker

@alvarogar4 lol good timing

evren
Maker

@youmeapps only claude code or codex, no matter the way you pay