Hey everyone. Just started testing OpenPipe for some internal metrics. Wanted to set up a basic cost monitoring dashboard.
Maybe I'm missing something, but I found the docs really hard to follow. 😅 The quickstart guide jumps from "install our agent" straight to high-level concepts without enough examples in between.
For instance, I spent an hour trying to figure out the exact label structure for their Prometheus exporter. Ended up having to guess by looking at the metrics in my Prometheus explorer.
```yaml
# Is this even right? The docs didn't show a full example.
openpipe_requests_total{project="my-app", endpoint="/api/chat"} 123
```
Did anyone else run into this? How did you get up to speed?
Not just you. Their docs prioritize flashy features over operational clarity.
> trying to figure out the exact label structure
You had to reverse-engineer from the metrics because they don't treat the exporter as a first-class component. It's an afterthought. For basic monitoring, you might be better off writing a small script to push metrics to your existing system. Adding another agent for this is rarely worth the complexity.
Your example looks right for a counter, but you'll need to check the metric type in their source if they don't document it.
Simplicity is the ultimate sophistication
You're definitely not the only one. This is a classic symptom of a product built by a team that's too close to their own code. They understand the internal abstractions so well that they forget what it's like to see the exporter for the first time.
Your guessed label structure is probably fine. The real issue is that you're having to architect around their documentation gaps, which adds hidden time and fragility to your setup. I've seen teams burn more hours reverse-engineering a tool's outputs than they'd spend building a simple custom metric collector from scratch. The complexity tax for these "integrated" monitoring agents is rarely justified for basic cost dashboards.
If the source code is your best documentation, you have to ask if the abstraction is even worth it.
keep it simple