Readable mode
Markdown that still feels like the source document
Markdown keeps headings, sections, tables, and line flow intact, which makes it useful for QA, handoff, and LLM context building.
Developer-first document extraction
OCR API
An OCR API that returns usable output, not raw text you still have to clean up.
LeapOCR starts from one upload surface, then lets you choose the result shape that actually fits the next step. Use markdown when a person or an LLM needs a clean document view. Use structured when a system needs fields it can trust immediately.
Why teams use this
Start from file upload or remote URL ingestion in the same OCR workflow.
Choose markdown for readable handoff or structured output for schema-fit extraction.
Add instructions, schema, templates, or bounding boxes only when the document actually needs them.
Upload experience
The parsing workflow stays simple on purpose: pick a source, choose the output shape, add optional guidance, and run.
Live request surface
Upload
Send a file or URL, choose output settings, and run OCR
OCR ready
SourceURL upload
Direct Upload
URL Upload
Document URL
https://storage.example.com/invoices/march/invoice.pdf
LeapOCR fetches the remote file and processes it the same way as a direct upload.
ConfigurationRequired
Model
standard-v2
Result format
structured
Processing instructionsOptional
|
Custom schemaOptional
invoice-schema.json
{ "invoice_number": "string", "invoice_date": "date", "total": "number"}Request
URL upload with model, format, instructions, and schema ready
Why it works
What you actually get
The key decision is not technical jargon. It is whether the next consumer needs a readable document, a structured object, or both with review context attached.
System-facing mode
Structured output shaped for downstream software
Structured mode returns a stable object instead of page prose, which means less post-processing, fewer brittle parsers, and cleaner integrations.
Layout data
Add bbox only when geometry matters
Bounding boxes are exposed as a boolean enhancement. Keep them off for pure parsing and on for review tools, overlays, and human-in-the-loop queues.
What you control
Choose the mode, then add control
These are the decisions teams actually make when they turn OCR into a production workflow instead of a raw text dump.
markdown
Mode · readable output
Return page content for humans and LLMs
Use markdown when the consumer needs a coherent page representation with headings, tables, and sections that still read naturally after OCR.
structured
Mode · schema-fit JSON
Return a stable object instead of page prose
Use structured mode when another system wants fields, arrays, and nested objects it can validate and write directly without another parsing pass.
instructions
Optional · string
Steer normalization and extraction choices
Use inline instructions for light behavior changes. Move heavier rules into templates once the workflow needs more room or more reuse.
schema
Structured only · object
Define the shape the downstream system can trust
Inline structured jobs need schema when you are not using a template. That is what keeps the result useful beyond a demo.
template_slug
Optional selector · string
Reuse a saved workflow setup
This is the cleanest path for production. It moves prompt, schema, and model choices out of every request body and into one reusable config.
extract_bounding_boxes
Optional · boolean
Attach geometry without changing the output mode
Bbox is not its own parsing mode. It is a toggle that adds page coordinates when the consuming workflow needs visual grounding.
Examples
Two common output patterns
Most teams settle into one of two patterns: readable markdown for human or LLM handoff, and structured extraction for systems that need direct writes.
Readable mode
Markdown output for review, handoff, and LLM context
This is the cleanest route when the next consumer still needs to read the page, quote it, or pass it into another model with the layout preserved in text form.
POST /ocr/uploads/url
json
{ "url": "https://example.com/claim-form.pdf", "file_name": "claim-form.pdf", "format": "markdown", "model": "standard-v1", "instructions": "Keep section headings and normalize dates.", "extract_bounding_boxes": false}
Result excerpt
json
{ "job_id": "job_01", "status": "completed", "result_format": "markdown", "pages": [ { "page_number": 1, "result": "# Claim form\n\n## Policy holder\n- Name: ..." } ]}
System-facing mode
Structured extraction with inline schema
This is the route for teams that need the API to return a stable object immediately instead of another layer of page text to parse.
POST /ocr/uploads/direct
json
{ "file_name": "purchase-order.pdf", "content_type": "application/pdf", "file_size": 918443, "format": "structured", "model": "standard-v1", "instructions": "Extract the order header and SKU table.", "schema": { "type": "object", "properties": { "po_number": { "type": "string" }, "order_date": { "type": "string" }, "items": { "type": "array", "items": { "type": "object", "properties": { "sku": { "type": "string" }, "quantity": { "type": "number" } } } } } }, "extract_bounding_boxes": true}
Result excerpt
json
{ "job_id": "job_02", "status": "completed", "result_format": "structured", "pages": [ { "page_number": 1, "result": { "po_number": "PO-10441", "order_date": "2026-03-10", "items": [ { "sku": "AX-44", "quantity": 12 } ] } } ]}
FAQ
Questions teams ask before wiring this up
Straight answers for teams evaluating how this workflow fits into production.
How do I choose between markdown and structured output?
Choose markdown when the result needs to stay readable. Choose structured when the next consumer is code, a queue, or a business system expecting named fields.
When should I bring templates into the parsing flow?
Use templates once a workflow repeats and the setup is stable. They let you keep model, instructions, schema, and review behavior in one reusable place.
Is bounding-box extraction a separate output mode?
No. It is an optional enhancement you can layer onto markdown or structured workflows when page geometry matters downstream.
Ready to test
Pick the output your workflow actually wants
Start with the output shape that matches the next step. When the workflow repeats, store the winning configuration as a template and stop rebuilding it.