Find candidates, broaden a short initial list, and verify each profile against the hiring criteria.
Candidate sourcing needs a broad pool and clear evidence. Exa Agent can interpret nuanced hiring criteria, research real people, and evaluate every requirement; Exa Deep Search can widen the pool when the initial Agent run returns too few candidates.
The primary path starts one live Exa Agent run and polls it outside the request lifecycle. If that run returns fewer than eight candidates, the app generates broader search angles, retrieves people with Deep Search, and sends those rows back to Agent for independent verification.
POST /agent/runs uses effort auto and the candidate schema. The schema preserves the normalized criteria and requires per-candidate evidence with met, not_met, or unknown decisions. The initial run returns up to ten candidates.
/agent/runs
When the first run has fewer than eight candidates, a planning model generates at least four complementary search angles. Each angle runs POST /search with type deep, category people, numResults 25, up to five additionalQueries, and a prompt asking for a broad candidate pool.
/search
Retrieved people are deduplicated, batched, passed through input.data, and checked by new effort-auto Agent runs against the original criteria. Only strongly verified candidates are added, up to a shortlist of ten.
/agent/runs
Find more creates a follow-up Agent run with previousRunId and passes existing or uploaded candidates through input.exclusion. The server caps exclusions at 300 and polling at 12 run IDs.
The demo does not use Deep Search as the primary candidate verifier. Agent owns the initial shortlist and every final criterion decision; Deep Search only broadens the candidate pool when the initial Agent result is small.