Question 1

Will it override my pinned model variant?

Accepted Answer

No when respectExplicitVariant is true. The prompt hook skips the Jev call when the session already pins model.variant, and the context hook never sets options on requests carrying a pinned variant.

Question 2

Does every tool-loop step pay for another Jev call?

Accepted Answer

No. The prompt hook stashes effort per session with replay-safe reuse, and tool-loop continuations apply that stash without another call. Only stash-less synthetic messages fall back to a one-off decision.

Question 3

What if my provider uses custom variant names?

Accepted Answer

It reads model list variants at startup and clamps abstract effort to the nearest supported variant, using id placement or effort strings in settings. Per-provider optionTemplates with effort placeholders cover remaining gaps.

Question 4

What happens when Jev is unreachable?

Accepted Answer

It fails open: missing keys, 8-second timeouts, or Jev errors leave model defaults untouched. Auxiliary title, summary, and compaction agents are excluded by default.

Question 5

How is it configured and authenticated?

Accepted Answer

Add it to the plugins array in opencode.jsonc with model, defaultEffort and maxEffort. It reads your existing Zen key from OPENCODE_API_KEY, or JEV_API_KEY as an override, and JEV_MODEL defaults to jev-1.13-free with jev-1.13 as the paid tier. The endpoint is resolved most-specific-first: plugin option, then JEV_ENDPOINT, then the default opencode.ai/zen/v1/systemone.

Question 6

How does an answer become a reasoning level?

Accepted Answer

Two questions go out together: a choice over minimal, low, medium, high and xhigh, and a high_stakes noul that at 0.7 or above bumps the level one step for irreversible, production, security, payments or migration work. A choice confidence under minConfidence, 0.35 by default, falls back to defaultEffort, and the result is clamped to maxEffort.

Question 7

How do I see what it decided?

Accepted Answer

The ./tui entrypoint auto-loads and shows the decided effort in the prompt footer, for example effort high/max. The server pushes each applied decision to it over plugin RPC through getDecision and decided, and the footer also pulls the current value on render so reconnects lose nothing. Config or option changes need the opencode process restarted, and the TUI restarted after an update.

Opencode Smart Reasoning

About this project

Highlights

Questions about this entry

More Jev resources like this

From the Jev community