Opening Models…

Jev playground

Explore an evaluation, then hand the setup instructions to your agent.

Edit a native TypeSafe request

Use the exact System One JSON, including model, state, and questions. This editor supports structured or omitted instructions and criteria. Close it to return to the form.

Model

1 Subject

Format

News alerts

Using Jev in a news alert system

Supply the topic, current evidence, and relevant earlier alerts together. Ask separate questions about relevance, new facts, and status changes. The same event can contain a new development.

Questions run together; each must stand on its own. Ask conditional questions upfront when the evidence is available, then use the relevant answers in your application. Keep timestamps, subscriptions, and delivery deduplication in code. Check generated headline support after the headline exists.

These probabilities are judgments, not verified facts or an alert policy. Compare against reviewed examples, including missed developments and duplicate alerts, before choosing thresholds. Additional questions can change token usage and cost.

Fan-out guidance ↗ · Cookbooks ↗ · Gateway OpenAPI ↗

2 Questions

Ask several questions about the same subject in one evaluation.

Each run sends your input to Jev.

Request JSON

Results

Not run

Run an evaluation to see results.

Response JSON

Use Jev with your AI agent

Create a token and copy the setup instructions to your agent.

Existing access tokens

Existing secrets cannot be shown again. Create another token if needed.

Usage

No evaluations in this period.

Includes signed-in evaluations and API calls, grouped by token. Cost uses reported input tokens and the configured estimate; it is not an invoice. Unmetered requests have no reported input usage, so totals may be incomplete. Failed requests before this view was added were not recorded. Unauthenticated requests are excluded; provider validation errors are included.