Download a PDF report
/v1/sessions/{id}/report.pdfScope exports:readScores, the AI summary and suggested follow-up questions, coverage with evidence, the decision and reviewers' comments. Add ?transcript=1 for the full transcript. Responds with application/pdf.
Export results as CSV
/v1/results.csvScope exports:readOne row per session, as text/csv. Every filter is optional; an invalid value is a 400 with issues:
| Parameter | Description |
|---|---|
interviewId | One interview's sessions; adds a score column per block |
status | created, in_progress, finished (run to its end) or ended_early (ended early by the candidate) |
decision | The team's decision: undecided, advance, hold or reject |
from, to | Sessions created within this range, each a full ISO 8601 date-time such as 2026-11-01T00:00:00Z (a date alone is a 400) |
cursor | The next part of a large export (from X-Next-Cursor), with the same filters |
Columns include when consent was given (consent_at), the scores, must-haves, the decision, the candidate's feedback (feedback_*) and the suggested follow-up questions. Then partial_result (yes) and blocks_not_analyzed (a count), filled when some blocks couldn't be analyzed, so the score covers only the others, and last candidate_id, to join the rows with your own records (an ATS, a bias audit's demographics).
A large export comes in parts, so each answers within the request timeout. A part that isn't the last has an X-Next-Cursor header and ends with a line starting # Truncated:. Ask again with cursor set to that value and the same filters until a part comes without it.