Jev in one minute
- Turn AI tasks into multiple-choice questions. Give Jev context and candidate answers. Choice questions return an option, probabilities and confidence. The source article reports responses in hundreds of milliseconds; your latency will depend on the request and network.
- Use it for frequent small decisions. Route requests, classify content, check whether summaries match a source, or assess agent tool risks without asking a generative model to write another answer each time.
- Let a coding agent handle the integration. Codex or Claude Code can install the TypeSafe skill, write the API calls and test your workflow. You define the task and review the results.
- Start alongside your existing workflow. Log predictions without changing live decisions. Set review thresholds using real examples and the cost of mistakes; send uncertain cases to a person or a larger model.

What does Jev do?
Think of Jev as a decision step inside software. Your code controls the workflow; Jev helps when a branch depends on the meaning of a message, an intent or a risk. Writing, complex reasoning and executing actions remain separate jobs.
It can select a category, judge whether a statement is true, or score an input against a rubric. A valid answer is not necessarily a correct answer.
Choose your next step
- Get started in four steps.
- See a complete worked example.
- Understand System One and the Jev name.
- Browse projects by use case.