CHOOSEAI infra
Jev Decision Engine
Jim Ma@Jim50836619
Jev acts as a decision engine, determining when to use an LLM for writing tasks, with user code controlling the execution gate.
TakeReplace the LLM call only where it was picking, rating or judging.
Jev cannot write a sentence. That part is true.
Four things the tutorials still get wrong:
→ Not multimodal. Text only — no image, audio, video.
→ Noul has no confidence field. One bare probability.
Wire a boolean into a confidence gate and that gate
does not exist.
→ Two context ceilings, not one. 64k per request, 32k
for state + your single longest question. Nothing is
reserved — but a big state plus one long question
hits 32k while you're nowhere near 64k.
→ 193.6x / 444.6x are TypeSafe's own workflow evals.
A vendor ceiling, not your workload.
Replace the LLM call only where it was picking, rating
or judging. Never where something has to be written.
Jev decides. The LLM writes. Your code holds the gate.
Free on Vercel AI Gateway until Sept 25. Six days to
run the same questions, Jev vs an LLM, on your own data.