Jev AI: One text. Many decisions.
Turn one piece of messy text into bounded answers your team and code can act on.
Which team should handle this ticket?
Does the customer need a response today?
How frustrated is the customer?
Go deeper when you need the model.
Learn how Jev handles typed decisions, compare workflows, and see the API contract after trying the tool.
Structured answers your code can inspect.
Jev reads the state once and evaluates every narrow question in parallel. Use the output to route, filter, review, or automate.
See how pricing worksTyped outputs
Numbers and labels your code can branch on, instead of another paragraph to parse.
Calibrated confidence
Every answer carries a probability so low-confidence cases can go to a human.
One call, many questions
Ask up to 64 questions about the same text and get one predictable response shape.
Simple, usage-based pricing
1 run = 1 credit, with up to 5 questions per run. Start with 50 free credits, then pick the volume you need. Cancel anytime.
Launch
$228For solo builders — 120,000 credits a year, 2 months free.
- 10,000 credits / month
- Online Playground
- Batch CSV evaluation
- Email support
Growth
$588For production teams — 1.2M credits a year, 2 months free.
- 100,000 credits / month
- Everything in Launch
- $0.49 per 1,000 credits
- Priority email support
Scale
$2,388For high-volume pipelines — 6M credits a year, 2 months free.
- 500,000 credits / month
- Everything in Growth
- $0.40 per 1,000 credits
- Fastest support response
New accounts get 50 credits, valid for 7 days. Enough to run the Playground examples and test a few of your own prompts. No card required.
Need more than 500,000 credits a month, an SLA, a dedicated account manager, or a custom integration? We will put together a plan for you.
Contact salesPrices in USD. Failed calls never consume credits. Full refund within 7 days if no credits were used.
Compare plans
Every plan includes the same Playground and typed outputs. You pay for volume and support.
| Feature | Launch | Growth | Scale | Enterprise |
|---|---|---|---|---|
| Monthly price | $19 | $49 | $199 | Custom |
| Credits (runs) per month | 10,000 | 100,000 | 500,000 | Custom |
| Price per 1,000 credits | $1.90 | $0.49 | $0.40 | Custom |
| Online Playground | ✓ | ✓ | ✓ | ✓ |
| Batch CSV evaluation | ✓ | ✓ | ✓ | ✓ |
| Support | Priority email | Fastest response | Dedicated manager | |
| Uptime SLA | — | — | — | ✓ |
| Custom integrations | — | — | — | ✓ |
Move from one text to a thousand rows.
Apply one decision rule to a file, inspect every row, and export results in CSV, TXT, or JSONL.
Open batch workspaceOne rule, every row, downloadable results.
What Jev AI is built to decide
Jev AI is a decision model for software that needs a bounded answer, not a paragraph. This guide separates the model, the workflow, the evidence on this page, and the limits to test before connecting an automated action.
Input
Text, structured state, or a record you already own.
Jev AI Model
Evaluates one state against questions you define, in one call.
Output
A typed choice, yes/no probability, or ordered score — your code can branch on it directly.
Not a chat replacement — a decision boundary.
A chat model is asked to explain, summarize, or draft. Jev AI is designed for a narrower software boundary: your code can branch on a typed value without first parsing a generated paragraph.
- Open-ended writing or brainstorming
- Long-form explanation, summaries, drafts
- Requires parsing free text to extract values
- No predictable output shape across calls
Good for: conversation, drafting
- Routing, triage, scoring, gates, verification
- Typed result your code can branch on
- Predictable shape: choice, boolean, score
- Up to 64 questions in one request
Good for: decisions with known answer shapes
Always keep application-side validation and error handling — Jev AI is a signal, not a guarantee.
Three question types, each with a predictable output shape.
A single Jev AI request can ask several narrow questions about the same state. Combine a route, a gate, and a review signal without forcing one broad prompt to carry every business rule.
| Question type | Output shape | Example | Best for |
|---|---|---|---|
| Choice | { label, probabilities, confidence } | "Route this to billing, technical, or sales?" | Support queues, intent routing, tool selection |
| Yes / No | { probability, confidence } | "Does this need a response today?" | Urgency gates, policy checks, escalation flags |
| Score | { value, confidence } | "How frustrated is this customer (1–5)?" | Severity, readiness, quality rubrics, review priority |
How a Jev AI request fits an application
Collect
Pull a de-identified ticket, message, review, or event record.
Define questions
Write each question and explain what every choice or score means.
Send & log
POST the state and questions; retain response fields with the decision log.
Validate & act
Route, filter, queue, or escalate — only after your own safety checks.
Jev AI owns the decision step. Your application still owns identity, permissions, retries, storage, and the final side effect.
Use confidence as a review signal.
Every Jev AI answer carries a confidence value. Set a threshold with representative examples and route low-confidence cases to human review instead of silently forcing every input into a queue.
Preview values on this page are illustrative until you run a request with your own key and data. Live output, local preview, and author-defined examples must remain visibly separate.
Use cases worth testing first.
Start with one repeated workflow where the labels and fallback path are easy to describe. Support-ticket routing is a useful first Jev AI use case — teams can name queues, define urgency, and compare the result with an existing human label.
Test Jev AI before production.
The batch workspace extends testing to CSV, TXT, and JSONL files so you can inspect a larger sample and export structured results.
Frequently asked questions
Can Jev AI replace a human review process?
No — Jev AI outputs should not be treated as an automatic guarantee. The right threshold depends on the cost of a wrong decision. A low-risk tag may tolerate automation; a refund, account restriction, or safety escalation should keep a human fallback.
How is this different from a prompt with examples?
A prompt with examples requires parsing generated text to extract values. Jev AI returns a typed result your code can branch on directly — no regex, no LLM parsing layer, no ambiguity about the output shape.
How many questions can I ask in one request?
Up to 5 questions per run, evaluated in parallel against the same input. One run can carry a route, a gate, and a review signal at the same time, for a single credit.
What happens when confidence is low?
When confidence falls below your chosen threshold, route the case to human review instead of applying an automated action. The confidence value is a probability estimate — treat it as a review signal, not a correctness guarantee.
Can I use this for batch processing?
Yes. The batch workspace supports CSV, TXT, and JSONL files so you can apply one decision rule to a file, inspect every row, and export results in a structured format.
Editorial scope: This is an independent Jev AI guide, not an official TypeSafe page. Content was reviewed on September 22, 2026 against the visible product flow and the linked API examples; preview values are not presented as measured production performance.