01 · The idea
What stayed with us
oMLX is a reference for the serving problems that appear on a single Mac: memory tiers, KV management, and concurrent requests. We keep it as a comparison point, not a claimed dependency.
02 · The local translation
What we did with it
oMLX is a useful comparison point for continuous batching and memory-aware serving on one Mac. The project helps us distinguish model capability from serving quality and ask when a warm shared model process changes throughput or response time.
03 · The boundary
Where the comparison stops
Our map names oMLX as a reference, not a measured head-to-head or a current production dependency. Its exact behavior and performance can change, so any adoption decision would require a pinned revision and same-workload comparison.