Demo·32 online · ≈1450 visitors
COMMUNITY OBSERVED

Exa Live Web Context for Jev

A live demo that compares Jev's typed answer with and without retrieved web evidence.

Overview & Result

Exa retrieves short, query-relevant highlights for a question, converts selected results into text entries in state.search_results, and asks Jev the same typed question with and without those entries. The demo exposes both answers and the request state.

How Jev fits in the loop

  1. Search Exa for six results with query-relevant highlights.
  2. Pass selected result titles, sources, dates, and highlights directly to Jev in state.search_results.
  3. Ask Jev the same bounded question again with an empty search_results array for comparison.
  4. Compute each source's displayed weight in demo code by rerunning Jev with that source removed.

How to reproduce

  1. Open the public Exa demo and select one of its provided questions.
  2. Compare the displayed Jev answers with and without web search for that question.
  3. Read the demo's How It Works page to inspect the Exa search request, the state.search_results input, and the Jev question.

Why this build matters

It shows a concrete way to ground a fast typed decision in fresh retrieved context while keeping the search and comparison steps inspectable.

Limitations

  • The public demo and its explanation are provided by Exa; JevForAgents has not independently benchmarked answer accuracy.
  • The examples are selected by the demo author and do not establish a general accuracy improvement.
  • The demo's source weights are computed by removing one result and rerunning Jev; they are not Jev-provided citations or a measure of source truth.

Patterns