How to Use Jev AI: Access, API and Pricing

To use Jev AI, start with TypeSafe’s console, obtain API access, and send a state plus typed questions to its System One endpoint. The request below shows how to classify a creative brief. It does not generate a video or run an autonomous agent.

Sources checked September 20, 2026. This guide uses official documentation. The example has not been executed against a paid API, and no response or latency measurement is presented as a test result.

New to the model? Read what Jev AI is and how System One differs from a generative model first.

1. Get access through TypeSafe

Open the TypeSafe console and follow the account’s access flow. The launch announcement described an early-access rollout; signing in does not by itself establish that your account can make API calls. Confirm your access and billing terms before integrating.

The official quick start covers the Playground and obtaining an API key from the dashboard. Try a small piece of text there, then move to a server-side request. Store the key in a server environment variable named TYPESAFE_API_KEY; do not put it in browser code, a public repository, or a NEXT_PUBLIC_ variable.

2. Send a small Choice request

After setting the environment variable in your terminal, run this illustrative request. Its structure follows the HTTP API reference; the creative brief and categories are our example.

curl --fail-with-body https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary '{
    "model": "jev-latest",
    "state": "I have a product photo and want a short clip for my shop.",
    "questions": {
      "workflow": {
        "type": "choice",
        "instructions": "Select the production workflow requested by this brief.",
        "criteria": {
          "image_to_video": "Animate a supplied image into a video",
          "text_to_video": "Create a video without a supplied image",
          "clarify": "The brief lacks enough information to choose"
        }
      }
    }
  }'

state contains the evidence. questions defines the evaluations. The workflow key lets your application find its answer; the actual question belongs in instructions, not just in that key.

Read the result under answers.workflow. For a Choice question, inspect choice, probabilities, and confidence. Keep the full response during development so you can inspect errors and token usage rather than assuming every request succeeds.

3. Decide what to do with the answer

The Choice documentation explains how option descriptions define the decision boundary. Our clarify option gives incomplete requests a useful destination. If the customer requests a video but never says whether an image exists, do not silently invent that asset.

Confidence is derived from the distribution of possible answers. It is not simply the selected option’s probability, nor a guarantee of correctness. Choose action thresholds using your own labeled examples; uncertain cases can go to a person or a clarification step. TypeSafe’s confidence guide explains the distinction.

For this experiment, print the proposed route before connecting it to a production workflow. Test a brief with a supplied photo, one with only a text idea, and one that leaves the required output unclear. Add contradictory briefs too. Record which ones need better instructions or an additional question.

Jev pricing: estimate input cost, then check usage

On the verification date, TypeSafe lists Jev 1.13 at $0.042 per million input tokens, with no charge for output tokens. The jev-latest alias points to jev-1.13.0. Check the current model and pricing reference before budgeting.

As an arithmetic example, 10,000 requests using 1,000 billed input tokens each would consume 10 million input tokens, costing $0.42 at that rate. This is an estimate, not a measured bill. It excludes retries, other models, infrastructure, and any separate provider charges.

Count the full request, including questions and option descriptions. Use returned usage to replace your estimate. Log the returned model version as well: an alias can change, while a versioned model ID makes later comparisons easier to reproduce.

Can you use Jev on OpenRouter?

OpenRouter has a Jev Latest listing that points to its current Jev family model. A listing alone does not confirm your account’s access, provider availability, or compatibility with an existing chat-completions client. Check that provider’s current API instructions and price before switching. The cURL example above targets TypeSafe directly and uses a TypeSafe key.

Is Jev open source? Which GitHub project should you use?

Distinguish model weights from integration code. TypeSafe links to a public System One adapter for Python, which implements its client interface using other LLM APIs. That repository is not a download of Jev’s model weights. We have not verified an official open-weight Jev release; use the documented hosted API path for this guide.

Before connecting a live workflow

Handle unsuccessful HTTP responses, set a request timeout in your application, and cap retries. Keep deterministic requirements such as asset existence and account permissions outside the model’s decision. Evaluate the cost of an incorrect route alongside request cost and latency.

If your goal is to create media today, browse ZNIX’s model directory separately. This tutorial describes an integration you could build; it does not mean ZNIX currently offers a Jev endpoint or playground.