A beginner guide
What is Jev?
Jev is TypeSafe AI's typed decision model: structured state goes in, and a constrained decision that software can validate comes out.
Why it exists
Some product decisions need a direct, bounded answer: which team should own a request, how risky is a transaction, or whether a person should review a case. Jev is designed for that decision boundary.
Jev versus a normal LLM
A normal LLM is typically used to understand or generate open-ended language. Jev is presented as a System One model for typed decisions. It can sit beside an LLM: one makes a bounded decision, while the other can explain it in prose.
Choice, Score and Noul
Choice picks one allowed option. Score returns a bounded numeric assessment. Noul answers yes or no with a probability. The application defines the allowed values and remains responsible for validation.
Real use cases
Teams can use typed decisions for ticket routing, fraud or risk triage, eligibility checks, prioritization and deciding when human review is required. Start with a mock boundary and test the decision against known cases.
System One versus System Two
TypeSafe AI uses System One to describe rapid, constrained decisions and System Two to describe slower, open-ended reasoning. The homepage comparison is a frontend simulation; its timing is not a universal benchmark.
Limitations and what Jev is not
Jev is not a general chatbot, a workflow engine or a guarantee that an output is correct. Typed output reduces ambiguity, but it does not remove the need for input checks, monitoring, policy controls and human oversight.
How to try it
Browse JevSkills to understand the ecosystem, then use TypeSafe AI's current documentation to choose a supported setup. Keep credentials on the server and validate returned values before a real action.
Best resources
Start with the JevSkills FAQ for direct answers, Explore for source-screened projects and Learn for primers. For product behavior and current integration details, consult TypeSafe AI's documentation directly.
TypeSafe AI: Introducing System One Models and Jev