Users need an integrated observability tool that can correlate LLM traces with traces and signals from other microservices. This would provide a complete view of the stack without requiring users to jump between multiple tools, addressing the current siloed approach.
LLM observability: Some companies claim you need a separate tool for it. But you don’t. This take might be controversial in some circles, but here’s why: - When you are deploying an LLM application, the LLM app is most likely a small part of your stack. And it’s usually interacting with other microservices, which are written in Java/Python/other languages and need standard observability. This creates silos. If you use LLM observability tools for your LLM apps, and more full-fledged observability for other parts of your application, you don't have a complete view of what's happening in your stack. - LLM o11y apps are also much more costly. If you scale by using them for all parts of your application, you will be spending a huge amount of money. - LLM tools are also not as feature rich, because many times they are still early in their product maturity. As a result, there are numerous stories about companies “outgrowing” single-purpose tools, because they needed cross-system tracing. (Eventually, they switched to classic distributed tracing, with LLM spans + metadata instead.) These are just some of the reasons an all-in-one tool for o11y is the right path. At SigNoz, we believe that for observability, there is a lot of value in a single pane of observability. That’s a principle we built our business on, and we’ve heard excellent feedback from others who have tried it our way instead. If you want to learn more about SigNoz, or LLM observability in general, I’d always love to chat.