Glossary: Intent
Intent is a closed-set label — usually a Choice — for “what kind of request is this?” so your code can pick a handler: deterministic lookup, specialist LLM, or human. It is not a generated plan, not a reserved API field, and not a TypeSafe enum shipped in the SDK. Jev (System One) answers the question you wrote; you map the winner to an action space.
This glossary entry is for builders who already heard of Jev / System One and need a citable definition before they treat intent as a magic key. jev.pro is an independent unofficial field guide — not affiliated with TypeSafe AI. Confirm the official intent-routing pattern on docs.typesafe.ai. We do not sell or issue API keys.
How it shows up
TypeSafe’s intent-routing pattern uses a Choice intent plus optional Score complexity. The word is yours; the API field is whatever key you choose in questions. Put type: "choice" and a map of at most 255 options. The answer includes choice, probabilities, and confidence. Low confidence means abstain, not “invent a new intent.” Gate irreversible handlers harder than recoverable ones.
A ranked route is what you build after the Choice returns: sort probabilities, apply a confidence threshold, then dispatch. Question IDs are not used in inference — write the full judgment in instructions. All questions in one POST /v1/systemone share the same state. Filter unused fields; jev-1.13 loses accuracy as distractors grow.
Schema-safe ≠ factually correct. A valid billing winner can still be a technical outage. Include an other option when “none of these” is a real department, then treat other as abstention unless policy says otherwise.
Contrast
Intent vs Choice. Choice is the primitive. Intent is a product role you assign to one Choice: routing among handlers. You can have Choices that are not intents (topic, language, severity band).
Intent vs Noul “is this billing?” Choice intent is relative among departments. A Noul asks whether billing is true in absolute terms. All Nouls can be low; a Choice still picks a winner among the set.
Intent vs ranked route. Intent is the label. Ranked route is the ordered handler list plus floors.
Intent vs verdict. The Choice key is a signal. The verdict is allow / hold / route / block after your policy.
Common mistakes
- Treating
intentas a reserved TypeSafe field that must be spelled exactly that way. - Omitting
otherand then forcing every ticket into a department. - Executing irreversible handlers at the same floor as “show a help article.”
- Asking Jev to generate a multi-step plan as the intent.
- Copying a Noul τ onto intent
confidence.
See also
FAQ
Is intent a reserved API field?
No. It is a question key you pick.
Intent vs Noul “is this billing?”
Choice intent is relative among departments. A Noul asks whether billing is true in absolute terms.
Should low-confidence intent invent a new label?
No. Abstain, hand off, or use other. Do not ask an LLM to expand the closed set in the typed slot.
Can I mix intent Choice and complexity Score?
Yes. Official pattern. Keep arithmetic in code; do not ask Jev to multiply weights.
Limits
This is a jev.pro working definition unless TypeSafe uses the same word as an API field. We do not invent search volume. Schema-safe answers are not automatically factually correct.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.