Run one query across many PDFs
Ask the same question of a stack of documents and get a row-per-file answer.
What this is
Have a folder of invoices, contracts, or reports and one question you'd like answered for each of them? A text query batch runs the same prompt against every file you pick, in parallel-ish batches, and gives you back a row-per-file answer you can export.
Good fits: pulling totals out of invoices, extracting dates from contracts, scoring résumés against a rubric, classifying documents, or any "do X for each of these" task.
Step by step
- Open your workspace and upload the files. If you haven't already, drop in the documents you want to query.
- Start a text query batch. From the workspace, choose the files you want to include and open the text query form.
- Write your question once. One prompt, phrased as if asking about a single file — Curie runs it against each file in turn. Be specific about the output format you want (e.g. "Respond with only the invoice total as a plain number, no currency symbol").
- Launch the batch. The batch starts running — three items at a time with a small delay between to stay friendly to upstream rate limits. Progress updates live.
- Review the results. Each row shows the file and the model's answer. Retry any individual row if the answer came back empty or malformed, or cancel the batch outright if you want to rework the prompt.
Good to know
- Batches are intentionally self-paced — throughput is traded for reliability on rate-limited upstream services.
- Each row is independent: the model doesn't remember what it said about file 1 when it's processing file 2.
- If you need cross-document synthesis, run the batch first and then ask a regular chat to summarize the results.