Description

we can cache token user slots per token

so we only need to detect them once and subsequent preparecalls can use the detected slot

Problem Statement

this will get rid of ~3 request per token

Implementation Hints

use a cache in the Chains object

User story

remove latency