Ask the help desk
The knowledge base is the policy handbook of a fictional company (expenses, travel, security, records, leave, IT, suppliers). Ask the same thing twice, in other words, and watch the second answer come from the cache.
Trace
The spans of the last request, on one time axis. The same spans go to Langfuse when its keys are set.
Since this instance started
- -requests
- -cache hit rate
- -p95 latency of a hit
- -p95 latency of a miss
- -list-price cost the cache avoided
- -misses answered after a failover
Percentiles over the requests this instance has served since it started; a free instance restarts when it sleeps.
Similarity lab
How similar does the cache think two questions are? No model is called: only the embedding model and the contrast guard.