Jev AI
Managed decision API for classifying, routing, scoring, and safety-checking text state, with a probability distribution and confidence score on every answer.
About Jev AI
Jev AI is a managed decision API: you send a state payload (state) and a set of named questions, and it returns structured answers with a probability distribution and confidence score, usable for classification, routing, scoring, and safety checks. Every answer comes with a probability and a confidence value, so you can set your own threshold to judge whether a result is trustworthy. A single request can contain up to 8 questions and cover classification, scoring, and safety checks for the same state in one call, which suits high-frequency decisions with several calls per second. Signup includes 5 free credits, so you can try it in the browser Playground first and then call the same endpoint from your code with an API key. Billing uses credits (credit): 1 credit covers 1,000 input tokens, rounded up to the actual token count with a minimum of 1 credit; output tokens are free, and failed requests are not charged. Credit packs are a one-time purchase, never expire, and there is no subscription or monthly reset. On data, the official docs state: requests are forwarded to the model to generate answers and content is not retained; only request metadata (time, status, credits, token count) is stored for usage history.
Best for
Best use cases
Classify, route, and score tickets, messages, or records, and decide whether to escalate to a human
Run safety checks or yes/no decisions inside a business process, with thresholds based on probability and confidence
Pay by usage in high-frequency scenarios, without maintaining a subscription or a self-hosted model
Compare before choosing
Per the official comparison page: Jev's model weights are proprietary and are called through the Optio endpoint, so no hardware of your own is needed; Kev is an Apache-2.0 open checkpoint (from 0.8B) and can run on Apple Silicon or your own GPU.
Per the official comparison page: Jev is billed by usage (1 credit ≈ 1,000 input tokens) and operations are handled by the host; once self-hosted Kev hardware is running, the marginal cost per call is close to zero, but the service, version management, and calibration checks are yours to run.
Per the official comparison page: Kev suits high and stable usage, isolated environments, or Mac-only teams; choose hosted Jev when you do not want to manage deployment and operations.
Features
Answers with probabilities
Every answer includes a probability distribution and confidence score, and you set the threshold yourself.
Multiple answers per call
A single request can include up to 8 questions, running classification, scoring, and safety checks on the same state at once.
Built for frequent calls
Designed for several decisions per second, where a typical request consumes 1 credit.
Playground and API
Try it in the browser, then call the same endpoint from your code with an API key.
Pros
- Answers carry a probability distribution and confidence score, and you can use them with thresholds.
- A single request can include up to 8 questions, running classification, scoring, and safety checks on the same state at once.
- Usage-based billing: credits are bought once, never expire, and there is no subscription or monthly reset.
- Output tokens are not billed, and failed requests are not charged.
- Signup includes 5 free credits, so you can try the browser Playground first.
Cons
- The model weights are proprietary and can only be called through the hosted endpoint, so you cannot self-deploy (per the official comparison page).
- Concurrency and requests per minute depend on the workspace plan, and exceeding them returns a 429 rate limit (per the official docs).
Pricing Plans
Free
- 5 credits
- 1 workspace
- 1 concurrent request
- Free Playground
Starter
- 100,000 credits
- About 100,000 typical requests
- Credits never expire
- 1 workspace
- 3 concurrent requests
- 120 requests per minute
- Up to 5 API keys per workspace
- Selection, scoring, and yes/no questions
- Probability and confidence on every answer
- Up to 8 questions per request
- Usage and request history
- Free Playground
- Email support
Pro
Team
* Pricing information is for reference only. Please refer to the latest announcement on Jev AI's official website.