Computer-Use Cache
A local OpenAI-compatible proxy that caches repeat model requests and can use Jev to judge safe semantic reuse after exact cache misses.
Added to Jevfast
How it uses Jev
After exact lookup misses, Jev compares full current request and candidate request/response context. It returns a candidate or none; configured confidence gating accepts reuse, while uncertainty falls through to the upstream model.
What Jev decides
One-candidate cache reuse requestExample answers · not a recorded Jev response · Source ↗
Question 1 · reuse
YOUR APP
INSTRUCTION
Which cached response can satisfy this request unchanged? Choose none if reuse is unsafe or uncertain.
STATE
Current request and one eligible cached request/response candidate keyed candidate_0. The candidate key and content are dynamic per cache state; the integration fixture demonstrates this representative one-candidate shape.
JEV · CHOICE
- none
- candidate_0
App workflow
- Receive request
An OpenAI-compatible client sends a request to the local proxy.
- Check exact cache
The proxy applies bypass and sensitivity checks, then attempts an exact eligible cache lookup.
- Judge near matches
On an eligible exact miss, Jev compares current request context with recent candidate cache entries.
- Reuse or forward
A confidently accepted unchanged response is returned; otherwise the proxy sends the request upstream and stores a successful response.
Why it is interesting
The safety criterion is unchanged reuse, with explicit refusal when new observation, execution, editing, or guessing would be needed.