Jev is TypeSafe AI's decision model: give it context and bounded questions, and it returns values that software can act on. It is useful when your application needs to choose, rate or check something. This guide explains the three question types with original examples, then shows a separate Dreamina workflow for illustrating the idea. Jev handles judgments; Dreamina handles visual creation.
What Is Jev AI?
TypeSafe introduced Jev on September 15, 2026, as its first public System One model. Its launch announcement describes an approach built for automation: evaluate a supplied situation and return a constrained decision. The announcement opened early access; consult the official site for current availability.
The input has two parts: state, meaning the information to evaluate, and questions, which define what to decide. A support ticket can be the state; the responsible team can be the question. The answer is a selected option, a score or a probability, rather than a drafted reply. TypeSafe's model introduction explains this machine-oriented interface.
A practical way to understand the distinction is to separate interpretation from action. A model may identify a ticket as a billing issue. Your application still decides whether to place it in a queue, request more information or wait for a person. Designing that second step is part of building the application, not something a structured answer automatically does for you.
Concept illustration: disordered information becomes bounded decisions. This AI-generated editorial artwork is not a Jev output or a TypeSafe interface screenshot.
Copy & Paste Jev Prompt Examples: Choice, Score and Noul
Use these original examples to design questions in the official Playground. Each has a sample state and a question definition; none is a recorded model response. Paste the state into its field and use the JSON as the questions object. For API access and authentication, follow TypeSafe's quick-start instructions.
Choice: Route a Support Request
Choose this primitive when the possible answers are categories. The result contains the selected option, probabilities across options and confidence. See the official Choice reference.
Sample state: My invoice lists two charges for one subscription. Please help me check the payment.
{
"support_team": {
"type": "choice",
"instructions": "Which team should review this request?",
"criteria": {
"payments": "Charges, invoices and payment questions",
"technical": "Broken features or access errors",
"other": "A request outside these categories"
}
}
}Implementation idea: add the selected team to a triage queue. Keep refund approval as a separate action, so a classification alone cannot trigger a payment change.
Score: Assess a Creative Brief's Specificity
Use Score for an ordered rubric. Its value can fall between levels, which start at zero; a three-level rubric therefore spans 0–2. The Score documentation also provides probabilities and confidence for interpreting the result.
Sample state: Create a wide editorial illustration of a glass prism sorting floating cards into three paths, on white with cyan accents.
{
"brief_specificity": {
"type": "score",
"instructions": "How concretely does this brief describe the intended scene?",
"criteria": [
"No identifiable subject or scene",
"A subject is stated but its presentation is vague",
"The subject, composition and visual treatment are clear"
]
}
}Implementation idea: use the result to prioritize briefs for review. This rubric concerns specificity, not artistic quality, factual accuracy or permission to use any depicted material.
Noul: Check Whether a Request Asks for an Edit
Noul evaluates a proposition and returns a value between 0 and 1. It has no separate confidence field. The official Noul reference explains this yes/no probability primitive.
Sample state: Keep the subject in my existing illustration, but replace the dark background with white.
{
"requests_edit": {
"type": "noul",
"instructions": "The request explicitly asks to modify an existing visual."
}
}Implementation idea: flag the request for a reference-image workflow. A positive result does not establish that the image was actually uploaded; your application should check whether the file exists.
Jev vs Generative AI: Choose by the Output You Need
Start with the deliverable. A category that selects the next software action is different from an image that explains an idea to a reader. These tools can belong to the same broader project without providing the same service.
For the illustration task, Dreamina's AI image generator provides the visual creation step. If your deliverable is a moving scene, explore its AI video generator. This is a comparison of roles, not a claim of a Jev integration or an automatic handoff between products.
How to Create Jev Explainer Visuals with Dreamina
For a blog cover or presentation, build the explanation first and illustrate one idea at a time. A visual metaphor can make the topic approachable, but technical labels and comparisons should come from the facts you have checked. The following manual workflow creates artwork about Jev, not output from Jev.
Step 1: Open the Image Generator
Open Dreamina and sign in. Choose AI Image to begin a still illustration. Decide whether the artwork will support a blog cover, a slide or a social post.
Step 2: Add a Visual Prompt and Choose Settings
Paste the visual prompt below, adding a reference image if you need one. Select GPT Image 2 when available and choose a landscape 16:9 ratio for this example. Keep the factual headline separate from the generated scene so you can check every word.
Step 3: Generate, Review and Download
Generate the image and compare the available results. Refine the prompt if the composition makes the decision paths difficult to distinguish. Download your chosen image and add it beside your verified explanation.
Copy & Paste a Jev Explainer Image Prompt
Create a wide editorial 3D illustration about turning unstructured information into bounded decisions. On a white studio background, show loose frosted-glass cards and cyan filaments on the left, a clear prism in the center, and three orderly geometric paths on the right. Use a selected cube, a stepped series of blocks and two contrasting discs as visual metaphors for choice, scoring and yes/no judgment. Add soft natural shadows, matte porcelain and restrained cyan accents. Keep a clean 16:9 composition with breathing room. No readable text, logos, watermark, product UI or humanoid robot. Make it a conceptual illustration, not a technical architecture diagram.
For a different composition, replace the prism with a transparent sorting tray or replace the studio with an overhead paper-collage scene. Preserve the distinction between loose inputs and bounded outputs. A coherent metaphor is more helpful here than a crowded montage of code, robots and fictional dashboards.
Before sharing, check the illustration at thumbnail size. The broad direction should remain legible without the headline. If you add arrows or labels later, ensure they describe your explanation accurately rather than suggesting that one question's answer is automatically fed into another in the same call.
Check the Decision Before You Automate
A valid answer format does not establish that a judgment is correct. TypeSafe's confidence guide describes a statistic derived from the probability distribution returned for Choice and Score. It helps expose uncertainty; it is not a universal success guarantee. Choose review thresholds using your own task and observed results.
- Define one decision at a time. Separating a brief's specificity from its originality makes an unexpected score easier to investigate.
- Include genuinely ambiguous examples in evaluation. A request that mixes a payment problem with a broken feature is a useful test case.
- Record the input, question definition and resulting action. That record lets you distinguish a model error from an application rule that needs revision.
- Keep a manual-review path. Decide what happens when information is absent before you decide when to automate.
- Verify numeric and visual claims independently. A persuasive explanation or attractive illustration should not become evidence for a benchmark claim.
TypeSafe's workflow patterns show ways to compose bounded judgments in code. For a first experiment, choose one low-impact decision with an observable outcome. Compare it with your current process, inspect disagreements, and expand only when the behavior is useful in that setting.
FAQs
Is Jev a Chatbot?
Jev is a decision model rather than a chatbot for free-form answers. You supply context and typed questions, then consume the structured results in an application. If the task is drafting an explanation or a reply, use a text-generation tool for that part.
Can Jev Generate Images or Videos?
The documented Jev interface returns decisions, not images or videos. For creative media, use a visual-generation tool such as Dreamina. A picture explaining Jev is an illustration about the model, not something the decision model generated.
Does Type-Safe Output Mean Jev Is Always Right?
No. Constraining the shape of an answer does not prove that the selected answer is correct. For example, a valid team label may still route a request to the wrong team. Evaluate the results against the task you actually need to perform.
Is Jev Available Inside Dreamina?
This guide does not describe a Jev integration in Dreamina. Use TypeSafe's official access route for Jev decisions and Dreamina for separate visual creation. The two workflows have different inputs, outputs and purposes.
Can I Try the Visual Workflow for Free?
Dreamina offers free credits for trying visual creation; check your account for the current allowance and selected model's requirements. That offer is separate from Jev access and API billing. Check TypeSafe directly for its current availability and pricing.
Turn Your Jev Explanation into a Visual Story
Choose the tool around the output: use a decision model for bounded judgments and a visual generator for artwork. Once your Jev explanation is fact-checked, pick one metaphor, generate a clear illustration and place it beside the relevant paragraph. Start with the landscape prompt above and adapt the materials, colors or composition to your audience.
Product information checked September 23, 2026. The decision-question examples are original illustrative inputs and were not executed against the Jev API.