Small decisions. Interesting possibilities.Submit contentSubmit

How it uses Jev

When optional Jev evaluation is enabled, the adapter sends bounded text from a distinct user request and the preceding turn to Jev. Jev judges whether remembered context could help. A positive result recommends Search to the coding agent; the agent decides whether to run it. Disabled or failed evaluation falls back to the existing reminder cadence.

What Jev decides

Should coding memory be searched?Example answers · not a recorded Jev response · Source ↗
Question 1 · search
YOUR APP
INSTRUCTION

Would searching Coding Memory likely materially help answer the current prompt, given the previous turn?

STATE

Caller-supplied conversation text, including the current prompt and previous turn.

JEV · NOUL
YesNo

App workflow

  1. Collect an eligible turn

    Deduplicate user requests and bound the current and previous-turn text.

  2. Ask Jev

    Evaluate whether memory context could help the current request.

  3. Guide the agent

    Recommend Search or skip; the agent performs any search, with existing reminder behavior as fallback.

Why it is interesting

The model is used only to decide when memory retrieval may be useful. The coding agent keeps control over the actual search, while deduplication and fallback prevent repeated or failed Jev calls from disrupting the session.