Sokudo
Request Lifecycle
First request warms it. Later requests benefit.
01
Cold request
Full prefix prefill. The system prompt, tools, and long context are computed.
0 cached tokens
02
Cache entry exists
Reusable KV state is retained under the provider’s scope and retention policy.
prefix warmed
03
Warm request
Matching prefix is loaded. Prefill starts only where the new suffix begins.
cache hit
A “partial hit” reuses the longest matching prefix, then computes the remainder.