Smart OCRv1.0Stop writing custom scripts to parse Textract BoundingBoxes. SmartOCR wraps AWS Textract (and other models) to give you clean, schema-validated Structured JSON instantly, saving you months of engineering time.
Smart OCR routes every document across a curated stack of AI providers and local models, then returns Structured JSON AI Models validated against your schema.
{
"invoiceNumber": "string",
"vendor": "string",
"currency": "string",
"lineItems": [
{
"itemName": "string",
"qty": "number",
"total": "number"
}
],
"totalDue": "number"
}{
"invoiceNumber": "INV-2847",
"vendor": "Lens & Light Co.",
"currency": "USD",
"lineItems": [
{ "itemName": "Sony A7 IV Body", "qty": 1, "total": 2498.00 },
{ "itemName": "24-70mm f/2.8 GM II", "qty": 1, "total": 2298.00 },
{ "itemName": "Aputure 600d LED", "qty": 2, "total": 3790.00 },
{ "itemName": "V-Mount Battery Kit", "qty": 4, "total": 1196.00 }
],
"totalDue": 9782.00
}Smart provider routing across industry-leading AI models including AWS Textract, Google Cloud Vision, and OpenAI. An AI-driven orchestration layer routes each document to the optimal engine — and automatically falls back across providers if any upstream degrades. No single point of failure in your extraction pipeline.
An LLM-backed normalization layer cleans, types, and reconciles every extracted field against your schema — fixing OCR noise, unifying date and currency formats, and returning structured values your services can trust out of the box.
A local ONNX model detects table geometry before routing, preserving row/column relationships for invoices, statements, and structured forms. Nested line items map cleanly into your typed JSON schema.
Bring your own JSON schema. Our Structured JSON AI Models map and validate fields against your contract — nested objects, typed values, required keys — so the response is safe to insert into your database without a post-processor.
PDFs containing embedded digital text are parsed natively in milliseconds, bypassing OCR entirely. Character-perfect text, lower latency, and no per-page OCR cost on the documents that don't need it.
Scoped API keys, per-key rate limiting, signed webhooks for async jobs, and audit logs on every request. Production controls without operational overhead.
Smart OCR is built for teams handling sensitive financial, medical, and legal documents. Privacy and isolation are defaults, not add-ons.
Documents are processed in memory and discarded from our servers immediately. Data routed to upstream AI providers is transmitted via secure APIs under strict enterprise agreements that prohibit model training on your data.
All traffic is encrypted in transit with TLS 1.2+. API keys are stored hashed at rest, requests are isolated per tenant, and provider credentials never leave our infrastructure.
Deploy the Smart OCR pipeline directly inside your own AWS or GCP environment. By utilizing your own cloud infrastructure and enterprise API keys, you maintain complete custody of your data from end-to-end, ensuring full compliance with HIPAA, GDPR, and SOC 2.
POST a file and a target schema to /v1/extract. Receive validated, type-safe JSON. No prompt engineering, no token math.
$ curl -X POST https://api.smartocr.dev/api/ocr/extract \
-H "x-api-key: YOUR_API_TOKEN" \
-F "file=@/path/to/invoice.pdf" \
-F 'schema={"invoiceNumber": "", "totalAmount": 0}'Create an account today and instantly receive 10 free API credits to extract data from your first 10 documents. No credit card required.
We're a small, dedicated team of AI and infrastructure engineers with backgrounds in document AI, distributed systems, and developer tooling. We started Smart OCR after spending years gluing together brittle, single-vendor OCR pipelines that failed at the worst possible moment.
Our mission is simple: build the most resilient multi-provider OCR fallback pipeline in the world, and expose it through an API a developer can integrate in an afternoon. No black boxes, no surprise lock-in, no marketing-grade accuracy numbers.
Every routing decision, confidence score, and provider hop is observable in your dashboard — so you can debug a failed extraction the same way you'd debug any other service in your stack.
No subscriptions. No surprises. Credits never expire.
10¢ per page
8¢ per page (Save 20%)
5¢ per page (Save 50%)