Jev Is Now Available on Emergent via Universal Key

Jev is now available on Emergent through the Universal Key. The typed decision model from TypeSafe AI lets builders make fast decisions on text inputs, such as classification, ranking, and routing, without a separate API key.

Jev makes typed decisions instead of generating text

Jev is not a chat model. Instead of writing sentences, it returns a typed answer with a calibrated confidence score. You give it some text and a question, and it replies with a set choice, a score, or a true-or-false call.

TypeSafe calls this class of model a System One model, a nod to fast, intuitive thinking. It fits the decision steps inside an app: sorting a support ticket by urgency, routing a message to the right team, scoring a lead, or checking whether a user input is safe to process.

The model reads text only. For images or audio, you first convert the asset to text or pair Jev with another model. Our Jev explainer covers how the model works in more detail.

Adding Jev takes no API key or manual setup

Because Jev runs through the Universal Key, you do not need a TypeSafe account or a separate key. Access and billing stay inside your Emergent project.

There are two ways to start. Describe your use case to the Emergent agent and it wires Jev into your build. Or open Manage, then Connectors, in your app dashboard and add Jev there.

TypeSafe reports major speed and cost gains on decision work

TypeSafe positions Jev as faster and cheaper than a general language model on decision tasks. Its launch announcement puts latency at less than 100 milliseconds and says Jev is up to 100 times faster and less expensive than other frontier models.

These are vendor figures. The launch showed TypeSafe's own comparison rather than an independently run benchmark, so expect real results to vary by task and setup. A typed answer is also not a guaranteed-correct one, so TypeSafe advises calibrating Jev on your own labeled data and using its confidence scores to decide when to act.

TypeSafe AI came out of stealth on September 15, 2026, with a $40 million seed round led by DCVC. Almeida previously worked on the reinforcement learning research behind ChatGPT at OpenAI. He founded the company in 2024 with Erik Gafni and Sasha Sheng.

Jev complements the LLMs you build with on Emergent

Jev does not replace the language models on Emergent. You still build with GPT, Claude, and Gemini for generation, reasoning, and conversation. Jev handles the smaller decisions those apps make along the way.

A support app might use a language model to draft a reply and use Jev to decide which queue the ticket belongs in. A content tool might generate a draft with one model and use Jev to flag anything that needs review. Both run in the same build.

This split is the point. Language models are strong at open-ended work. Jev is built for the fast, repeatable calls that sit between those steps.

What this means for Emergent builders

Jev gives Emergent builders a purpose-built tool for the decision points in their apps, available without extra setup. You describe the job, and the agent handles the wiring.

Start Building on Emergent and add Jev to your next project today.