Release - 2023-05-25
Data Retrieval Observability
New observability features for Metal.
by Taylor Lowe
Vic Rattlehead learning about observability
As the number of applications deployed on Metal continues to grow, so too has the need for improved visibility into application performance and user behaviors. Today we’re releasing observability enhancements to Metal that address these needs.
This release focuses on the heart of Metal’s platform: indexing and querying data. We now expose every query made against an index as a log. These logs are conveniently accessible in Metal’s UI.
Logs of queries made against an index in Metal
We’re also releasing a dashboard to keep track of queries and cosine similarity distances.
Dashboard to keep track of queries and cosine similarity distances
There are a few immediate benefits.
To start, it’s always good to run a sanity check. These logs can be used to verify that your application is successfully querying data in an index.
Further, having a history of user queries will help you understand what kind of information your users are looking for. Are they searching for specific answers or using certain keywords more frequently? Does your index contain the correct data based on the queries made against it?
Lastly, observing how your users query an application can better inform the chunking strategy for your data. A chunking strategy may have seemed good in theory, but the rubber always meets the road with real world usage.
As LLMs powered applications continue to increase in scope and and use, observability into their behaviors – and the users who interact with them – will be crucial. These logs will give developers using Metal a clearer insight into current behaviors and usage, but please join our Discord or drop us a line if there are additional improvements you would like to see.