I implemented JEV @typesafeai yesterday and can say it is a new complementary layer.

Nanda@thepeekpoker

Sep 20

SignalIf they require parameters, JEV has already selected the tool and the other model fills them in; because part of the decision is already solved, this also takes less time.

I implemented JEV @typesafeai yesterday and can say it is a new complementary layer. One example: Before, an X%-scored small model handled the tool calls, taking 500–1700 ms on average. Now the tool calls go through JEV first. If they have no parameters, it is a direct win in speed and cost (50–500 ms). If they require parameters, JEV has already selected the tool and the other model fills them in; because part of the decision is already solved, this also takes less time. So it is a win. You can also have JEV set tool-call parameters when they are booleans or enums.

Original post