Lovable: Add Typed Decision Models With Jev
LovableLovable's AI features can now run on Jev Latest, a typed decision model from TypeSafe that returns a category, a score, or a yes/no probability instead of generated text. The addition gives builders a faster, cheaper option for tasks like ticket triage, content moderation, routing, and search ranking, where a structured answer is more useful than a paragraph. Builders can turn it on simply by asking Lovable to use it when setting up an AI feature, with no separate TypeSafe account required.
Key Takeaways
- Jev Latest returns typed answers, such as a category, a score, or a yes/no probability, instead of generated text.
- It is built by TypeSafe as a "System One" decision model, distinct from the generative language models Lovable's AI features already support.
- Good fits include ticket triage, content moderation, routing, and search ranking, tasks where a structured value matters more than prose.
- Builders can turn it on just by asking Lovable, with no separate TypeSafe account required.
- Jev is part of a broader industry trend toward faster, cheaper specialized decision models as an alternative to general-purpose LLMs for narrow tasks.
- The addition expands the range of AI features available inside Lovable-built apps beyond text generation and chat.
A Model Built for Decisions, Not Text
Most AI features inside Lovable apps run on large language models that generate prose. Lovable's AI features now offer an alternative: Jev Latest, a typed decision model built by TypeSafe that skips text generation entirely and returns a typed, structured answer instead, such as a category chosen from a fixed set, a score against a rubric, or a calibrated yes/no probability.
Where Jev Fits
Because Jev returns structured output rather than prose, it suits tasks where an app needs to act on the answer directly rather than display it to a person. Lovable points to ticket triage, content moderation, request routing, and search ranking as good fits, cases where a traditional generative model would otherwise need extra parsing and prompt engineering just to get a usable value back.
How to Use It
Builders do not need a separate TypeSafe account to try Jev. They can simply ask Lovable to use it when setting up an AI feature, and Lovable wires it into the app's AI features configuration alongside the generative models already available.