Jev AI

Jev AI use cases

Retrieval & knowledge

Bring outside knowledge into a decision: search for evidence, then let Jev weigh it.

Retrieval & knowledge

Bring outside knowledge into a decision: search for evidence, then let Jev weigh it.

About retrieval & knowledge

Jev decides from the state you send; it does not browse or remember documents. Retrieval is how outside knowledge gets into that state, and the quality of a decision depends on what was retrieved.

These pages cover both ends of that. Check the passages your own retriever returns before they reach a model, or run a live web search and see how the evidence changes Jev’s answer. In both, the retrieved text is treated as data, never as instructions.

Which page fits your task

  • RAG EvaluationYou run your own retrieval (vector or keyword search) and want to check or rerank each retrieved passage before generation.
  • Live Web ContextYour question depends on recent public facts and you want Jev to answer with live web search results as evidence.

Common mistakes

  • Accepting passages because they match the topic, when they do not answer the question.
  • Mixing document versions, so an outdated passage outranks the current one.
  • Passing retrieved text to a model as if it were trusted instructions.

Retrieval & knowledge FAQ

Does Jev search the web or my documents by itself?

No. Jev reads the state you send. The Live Web Context page runs a web search first and passes the results in; for your own documents, your retriever supplies the passages.

Can Jev replace my vector database?

No. Keep your search for recall across the whole corpus. Use Jev on the shortlist it returns to filter, check and reorder passages.

Further reading · reviewed 2026-09-23

  • Classifying RAG passagesTypeSafe documentation · docs.typesafe.ai/cookbooks/classifying_rag_passages
  • Re-ranking a search shortlistTypeSafe documentation · docs.typesafe.ai/cookbooks/rerank_typesafe

One decision model. Your own rules.

Use a Noul for a yes/no probability, a Choice for a fixed set of outcomes, or a Score for an ordered rubric. Ask focused questions about the same input and combine the results in code.