What is Jev?

1 of 8

How does Jev work?
One state, typed questions, structured answers
State: text, JSON, or arrays of text.
Questions: Choice, Score or Noul.
Jev: evaluates in one pass.
Results: typed answers and probabilities.
Your code: act, review, or escalate.
You define the answer types in advance.
Jev returns structured outputs instead of prose.
Your application still owns the rules and the final action.
Noul is TypeSafe's yes-or-no probability primitive.
Jev is a decision engine inside a larger software workflow.

Why use Jev?
Why businesses care about it
It returns structured outputs your software can use directly.
It removes much of the parsing and formatting work common with chat-model outputs.
It can answer multiple independent questions about the same state.
It includes probabilities and confidence signals for decision-making.
It fits repeatable software decisions better than open-ended generation.
App state, then Jev, then route, score or review.
Jev is useful when your application needs a defined decision, not a paragraph.
Think: ticket routing, review gates, classification, and software workflow checks.

What is Jev good at?
Common fits for a System One model
Ticket routing: Choose the right team or queue.
Risk or fraud triage: Score or flag what needs review.
Agent next step: Help decide whether to call a tool, retry, or stop.
Context selection: Pick the most relevant context or evidence.
Review before action: Check a proposed action before it runs.
Classification and scoring: Sort, label, and rate inputs on a rubric.
Best fit: when the answer space is defined in advance.
Jev shines when your software needs a fast, structured judgment.

Jev vs. an LLM
Different tools for different jobs
Jev: Returns typed decisions. Best for routing, scoring, and verification. Software can use the result directly. Provides probabilities and confidence. No paragraph to parse.
LLM: Generates text, code, and explanations. Best for writing, reasoning, and conversation. Humans usually read the output. Flexible, but less structured. Often needs parsing for automation.
Decision, then generation.
Use Jev for machine decisions. Use an LLM for human-facing output.

Where does Jev fit?
Jev, agents, MCP, APIs, and real systems
App or agent, then Jev decision, then app logic, then MCP, API, database or tools, then action.
Jev is the decision layer, not the action layer.
APIs and MCP connections are how software reaches tools and data.
Your code applies permissions, thresholds, and business rules.
Low-confidence cases can be sent to a person or a larger model.
Jev can help an agent decide what to do next, but other systems still perform the work.
Jev helps decide. Your systems still do the work.

When not to use Jev
When an LLM or other tool is the better fit
Do not use Jev when you need writing, conversation, or polished prose.
It is not built for open-ended brainstorming or research.
It is not the right tool for long explanations or complex generated output.
It currently works with text-based state, not images, audio, or video.
If your goal is generation, use an LLM instead.
Jev decides, then an LLM writes or explains.
Jev helps software make judgments. It does not replace every AI model.
Choose the model based on the job, not the hype.

Should you use Jev?
A simple decision guide
Do you need a reply, explanation, or generated content? Use an LLM.
Do you need a defined category, score, or yes-no estimate? Jev may fit.
Do you need both decision-making and writing? Use Jev for routing or scoring, then use an LLM for generation.
Do you need real-world execution? Pair the decision with code, APIs, MCP, or human review.
Good Jev use cases: routing, scoring, verification.
Not a Jev job: writing, open-ended research, long explanations.
Use Jev when the software needs a structured judgment, not a paragraph.

Lesson done.

Up next

Can AI answer your phone?

Where does this fit your business? Get your AI plan

Swipe to see each slide.

The short answer

Jev is TypeSafe AI's decision model: instead of writing text, it answers questions your software asks about a piece of text with typed answers, a choice, a score or a yes-or-no, each with a probability. It suits repeatable decisions inside software such as routing tickets or flagging work for review, while writing and conversation stay with a language model.

Three things to remember

01Jev returns a choice, a score or a yes-or-no with a probability, never a paragraph.
02It fits defined decisions like routing, scoring and checks, with your code owning the final action.
03For replies, explanations or research use a language model; low-confidence cases go to a person.

Every slide, in words

01
What is Jev?
A System One model for software decisions
Jev is TypeSafe's flagship model.
It is built for fast, structured decisions that software can use directly.
Instead of generating a paragraph, Jev returns typed answers and probabilities.
It is designed for automation inside software, not for chat-style replies to people.
Input or state, then Jev, then a typed decision: choice, score, probability.
Think: routing, scoring, classification, and verification.
02
How does Jev work?
One state, typed questions, structured answers
State: text, JSON, or arrays of text.
Questions: Choice, Score or Noul.
Jev: evaluates in one pass.
Results: typed answers and probabilities.
Your code: act, review, or escalate.
You define the answer types in advance.
Jev returns structured outputs instead of prose.
Your application still owns the rules and the final action.
Noul is TypeSafe's yes-or-no probability primitive.
Jev is a decision engine inside a larger software workflow.
03
Why use Jev?
Why businesses care about it
It returns structured outputs your software can use directly.
It removes much of the parsing and formatting work common with chat-model outputs.
It can answer multiple independent questions about the same state.
It includes probabilities and confidence signals for decision-making.
It fits repeatable software decisions better than open-ended generation.
App state, then Jev, then route, score or review.
Jev is useful when your application needs a defined decision, not a paragraph.
Think: ticket routing, review gates, classification, and software workflow checks.
04
What is Jev good at?
Common fits for a System One model
Ticket routing: Choose the right team or queue.
Risk or fraud triage: Score or flag what needs review.
Agent next step: Help decide whether to call a tool, retry, or stop.
Context selection: Pick the most relevant context or evidence.
Review before action: Check a proposed action before it runs.
Classification and scoring: Sort, label, and rate inputs on a rubric.
Best fit: when the answer space is defined in advance.
Jev shines when your software needs a fast, structured judgment.
05
Jev vs. an LLM
Different tools for different jobs
Jev: Returns typed decisions. Best for routing, scoring, and verification. Software can use the result directly. Provides probabilities and confidence. No paragraph to parse.
LLM: Generates text, code, and explanations. Best for writing, reasoning, and conversation. Humans usually read the output. Flexible, but less structured. Often needs parsing for automation.
Decision, then generation.
Use Jev for machine decisions. Use an LLM for human-facing output.
06
Where does Jev fit?
Jev, agents, MCP, APIs, and real systems
App or agent, then Jev decision, then app logic, then MCP, API, database or tools, then action.
Jev is the decision layer, not the action layer.
APIs and MCP connections are how software reaches tools and data.
Your code applies permissions, thresholds, and business rules.
Low-confidence cases can be sent to a person or a larger model.
Jev can help an agent decide what to do next, but other systems still perform the work.
Jev helps decide. Your systems still do the work.
07
When not to use Jev
When an LLM or other tool is the better fit
Do not use Jev when you need writing, conversation, or polished prose.
It is not built for open-ended brainstorming or research.
It is not the right tool for long explanations or complex generated output.
It currently works with text-based state, not images, audio, or video.
If your goal is generation, use an LLM instead.
Jev decides, then an LLM writes or explains.
Jev helps software make judgments. It does not replace every AI model.
Choose the model based on the job, not the hype.
08
Should you use Jev?
A simple decision guide
Do you need a reply, explanation, or generated content? Use an LLM.
Do you need a defined category, score, or yes-no estimate? Jev may fit.
Do you need both decision-making and writing? Use Jev for routing or scoring, then use an LLM for generation.
Do you need real-world execution? Pair the decision with code, APIs, MCP, or human review.
Good Jev use cases: routing, scoring, verification.
Not a Jev job: writing, open-ended research, long explanations.
Use Jev when the software needs a structured judgment, not a paragraph.

More in part 6

Next step

Where does this fit your business?

Six quick questions about your business, then your AI plan: what to start with, what a person still checks, and the lessons that go with it. Or bring it straight to a free thirty-minute call.

Get your AI plan Book a free call