Document Extraction
Upload a document, define a schema, and extract structured data with Sparrow.
Input document
PDF or image · max 5 MB · removed after inference
Drop a document or browse files
Supported: PDF, PNG, JPG, TIFF · max 5 MB
Extraction schema
JSON describing the fields to extract
Provide a JSON array of fields with their expected types. Sparrow returns a matching JSON document.
Processing options
Control how the model interprets your document
Table-only extraction
Focus on tabular content — best for dense tables, financial reports, lab results, portfolio statements.
Schema validation
Validate extracted values against the schema types before returning.
Standard works well for most documents. Advanced is recommended for complex forms.
Without a key, usage is limited to 30 calls per 6 hours and 3-page documents. Contact us for a key.
Try an example
Pre-loaded sample documents
Privacy by default
Documents are never stored — the upload is removed as soon as inference completes.
Response
JSON output from Sparrow
No results yet
Upload a document and define a schema, then run extraction to see structured output here.
Document Extraction
Structured data from invoices, receipts, statements, and tables — multi-page PDF, page classification, bounding-box annotation, and schema validation. No cloud dependencies.
Business Rules
Push formatting, derived fields, classification, and transformations to the LLM itself — no post-processing code, just typed schema fields with optional defaults.
Sparrow Agent
Orchestrate Vision LLM extraction with Text LLM reasoning — chain classification, extraction, and field validation with visual monitoring and error handling.