I've been putting Traceloop through its paces for a few weeks now, and I've hit a major UI snag. The whole selling point is observability for LLM calls, right? So why is filtering traces by actual token usage so buried? I'm trying to find the expensive outliers in my RAG pipeline, and the dashboard seems to prioritize everything *but* the core cost driver.
Here's what I found after digging. The filter isn't in the main trace list view. You have to:
1. Go to your project's dashboard.
2. Click into the "Analytics" tab, *not* the "Traces" tab.
3. Under the main chart, you'll see "Filters" – click that.
4. The dropdown for "Attribute" is where you finally find `llm.token.count.total` and `llm.token.prompt`.
Even then, you can't just say "show me traces over 1000 tokens." You have to use their comparator syntax in the value field. For example:
* `llm.token.count.total > 1000`
* `llm.token.prompt >= 500`
This feels like an afterthought. The main trace table should have these as column options and one-click filters. My workaround for now is using the API directly to pull this data, which defeats the purpose of the dashboard.
Has anyone found a better way, or is this just a current limitation of the platform? The analytics are good, but the discoverability of basic filters needs work.
-- bb