GPT Researcher
Research agent that scores scraped passages with Jev before drafting cited reports.
Added to Jevfast
How it uses Jev
When context filtering is configured as Jev or auto with a TypeSafe key, GPT Researcher splits scraped pages into text chunks and asks Jev to score each chunk's usefulness for the current research sub-query. It sorts by score, keeps chunks scoring at least 1.5 by default, and passes up to the result budget to the report writer. Jev calls are parallelized per chunk and across concurrent sub-queries; small page sets bypass filtering. Missing credentials or call failures fall back to local keyword ranking.
What Jev decides
Score passage usefulness for a research sub-queryExample answers · not a recorded Jev response · Source ↗
Question 1 · usefulness
YOUR APP
INSTRUCTION
How useful is this passage for answering the question: <sub-query>.
STATE
One nonempty text chunk from a scraped page, default 1,000 characters, judged against the current research sub-query.
JEV · SCORE
0
App workflow
- Plan and search
The research run produces sub-queries, searches for pages, and scrapes the results.
- Filter page chunks
When Jev filtering is enabled and the content is large enough, score each chunk in parallel for usefulness to its sub-query.
- Select context
Sort by score, keep chunks above the threshold up to the context budget, or use keyword ranking if Jev is unavailable.
- Write the report
Join selected context from sub-queries and send it to the report-writing model.
Why it is interesting
It creates a distinct usefulness-filtering stage between web retrieval and report writing: Jev scores bounded passage chunks so the writer sees fewer merely on-topic excerpts.