How can an AI that "can't chat" be played with? Developers use Jev to take over Agent for rapid decision-making.

Beating AI News Flash: TypeSafe, founded by former OpenAI researcher Diogo Almeida, just released its new model Jev yesterday. It does not generate text, only returns options, scores, and probabilities. A wave of open-source projects quickly emerged in the community, starting to plug Jev into Agents and software workflows.

Browser Use's jev-ultrafast is the most intuitive. At every step the browser takes, Jev determines from the current page "what to do next and which element to operate on," and only calls a small language model when text needs to be filled in. The team used it to search for flights from Zurich to London on Google Flights, and a complete run took only 7.1 seconds.

Jev Codex Router lets Jev first assess the difficulty of each round of a coding task, then decide which tier of model to hand it to; jev-mcp turns it into a tool for Agent fact-checking, Prompt Injection detection, and semantic ranking.

These projects share one obvious commonality: no one uses Jev as a chatbot. Instead, they plug it into the execution chains of software and Agents, specifically handling those high-frequency, limited-option judgments.