Jev from TypeSafe AI is now running inside Modus. We've been testing it since launch, and the results blew past what we expected, so it's rolling out across the platform, on Robinhood Chain, Solana, and every EVM chain we support.

First place it landed: tool routing. Every agent turn decides which tools to load and which prompt template to use, and an LLM used to make that call, slowly and often wrong. Jev now makes it in ~0.38s with calibrated probabilities, and routing precision nearly doubled, from 44% to 83%.

Second place: design references. Jev ranks the candidates and picks the best fit for the screen being designed, and usable first picks went from 11% to 66%, a 6x jump.

More of Modus is moving to Jev, and we can't wait to see what the community builds with it.

Benchmarks below

https://x.com/modusbuilders/status/2102310721144897737?s=20