What is Jev, the AI model that refuses to write sentences?

Jev is a software decision model: it returns bounded values and calibrated probabilities instead of conversational text.

Jev is not another chatbot. TypeSafe AI’s model accepts a piece of software state and returns constrained choices, numerical scores, probabilities, and confidence values that programs can use directly. Founder and CEO Diogo Almeida and several teammates previously worked on OpenAI’s RLHF, InstructGPT, and ChatGPT efforts; their San Francisco company, founded in 2024, opened limited early access on September 15 after a $40 million DCVC-led seed round and a reported valuation near $200 million.

TypeSafe calls Jev a “System One” model: fast, bounded judgment rather than open-ended prose. That narrower output can make automation easier to verify, but confidence is not the same as correctness—software should still escalate uncertain or high-impact decisions to people.