How to get started with Jev

Start without code

You can try your own text and question in this site’s independent free demo without supplying an API key. Choose a Yes/No rule or define 2–6 choices. To connect Jev to your own application, use TypeSafe’s separate official account and API access path below.

The demo allows up to 10,000 characters in total across text, question, rules and options, counted as Unicode code points. This is the site’s allowance, not Jev’s context window. Live checks may require human verification and can pause at usage or budget limits. Loading a template does not run it or supply a model result.

Open the custom editor · Load the feedback template

1. Start with the right access path

As of September 17, 2026, TypeSafe’s public site presents a waitlist. The official quickstart provides instructions for the Playground and API. An accessible documentation page does not mean every account has model access.

• If you do not have access: use the waitlist on TypeSafe’s official website. This site cannot approve accounts or issue API keys.
• If your account has access: open the TypeSafe console and follow its Playground or API-key flow.

Account availability and onboarding may change. The official console is the place to check your own status.

2. Try one narrow question

In the Playground, use a short fictional message as the input and ask one Noul question. For example, use a message requesting a callback and ask whether a callback is explicitly requested. Then change the input to a vague request for help and observe whether the result changes.

This is an exploration exercise, not an accuracy test. Starting with two contrasting examples helps you see whether your question describes the distinction you meant to make.

Jev calls the supplied information state. It can be a string, a JSON object, or an array. For richer tasks, use named fields to keep the message, supporting records, and relevant policy distinguishable. Input formats

curl --fail-with-body --silent --show-error \
  https://api.typesafe.ai/v1/systemone \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @- <<'JSON'
{
  "model": "jev-latest",
  "state": "Please call me tomorrow morning to discuss the setup.",
  "questions": {
    "callback_requested": {
      "type": "noul",
      "instructions": "Does the sender explicitly ask for a phone call?"
    }
  }
}
JSON

4. Add structure only when the task needs it

Use Choice when the next step depends on a category, Score when you have meaningful ordered levels, and Noul when you need a yes-or-no probability. Add independent questions about the same state together; handle their results in your application.

For a concrete next step, follow the support-ticket example. It shows how a documented output becomes a routing proposal.

5. Troubleshoot access and the request separately

If the console login does not finish, reopen the official console from its normal entry point and check that you are using the account with access. Avoid repeatedly changing your API request to solve a browser login problem. A successful login also does not establish that an API key has been issued. If you need account help, use TypeSafe’s official support path; never include a key or a login link in a public report.

For an HTTP request, inspect the status and error body before changing the question:

These codes are documented in the API reference. For 429 or 529, use exponential backoff. SDK defaults already retry; account for those attempts before adding an outer retry loop.

A timeout is different from a negative answer. Set an overall deadline and a bounded number of attempts, then route unresolved work for review. Ensure that retrying the model request cannot repeat a downstream action such as creating an assignment.

6. Troubleshoot a surprising answer

These are editorial debugging steps. The Score guide explains fractional values; our limitations page links the current vendor caveats and the separately labeled community observations.

The jev-latest alias can move to a newer model. Record the returned model identifier while testing; when reproducibility matters, use a versioned identifier supported by your account. Model aliases

7. Decide whether the first workflow is ready

Before enabling automatic actions, keep a small set of expected decisions, review mistakes, and choose thresholds for your workflow. Use the pricing calculator to estimate input-token spend and the comparison guide to decide what should remain with other tools.