Agent OCR
An agent reads the document in passes, extracting each field and re-checking the weak ones.
About Agent OCR & FAQ
Self-checking field extraction
Agent OCR is for documents where a wrong field matters. An AI agent extracts each field with a confidence score, then re-reads the low-confidence regions until every field meets your threshold or the run budget is used up.
- Default confidence threshold: 80%.
- Up to 5 passes per run by default.
- Each step is logged, so you can see how a value was reached.
When should I use Agent OCR instead of Simple OCR?
Use it for forms, invoices and IDs where each field has to be right. It makes more Gemini calls than a single pass, so it is slower and uses more of your quota.