NewOpenCodex
A provider proxy for coding agents with optional Jev model and effort routing.
Added to Jevfast
NewHow it uses Jev
The optional route decision builds Jev state from the task, signals and recent step evidence. It enumerates configured model-and-effort pairs as Choice option IDs and asks Jev which pair fits the next call.
What Jev decides
Next model routeExample answers · not a recorded Jev response · Source ↗
Question 1 · route
YOUR APP
INSTRUCTION
Which configured target and reasoning effort together fit the next model call?
STATE
Task, signals, recent tool evidence and the configured model candidates in the pinned test fixture: GPT-6 Astra at medium/high effort and GPT-5.6 Sol at low effort.
JEV · CHOICE
- openai/gpt-6-astra:medium
- openai/gpt-6-astra:high
- openai/gpt-5.6-sol:low
App workflow
- Prepare routing evidence
Collect the task, available signals, recent tool evidence and configured candidate model profiles.
- Ask Jev to route
Build a Choice criterion for each configured model-and-effort pair and submit one decision.
- Use the selected route
Validate that the returned Choice ID is in the configured option map, then apply that target and effort.
Why it is interesting
It uses Jev to pick both a target model and reasoning effort from a constrained configured set, rather than asking a language model to generate an unconstrained route.