Changelog

​

MCP Access and Reusable Extraction Workflows

• Connect AI assistants through the hosted MCP server — use OAuth 2.1 sign-in for Claude and other MCP clients, with built-in document extraction and exploration tools.
• Reuse existing uploads by record_id — rerun sync, async, streaming, or batch extraction on previously uploaded files instead of sending the same file again.
• Upload files in one request via /api/v1/upload/file — send multipart form data and get a record_id back immediately for follow-on extraction calls.
• Embed metadata directly in JSON results — use metadata_options=unified to merge bounding boxes and confidence data into the response structure you consume.

​

AI-Powered Schema Generation from Natural Language

• Natural language input — describe what you want to extract in plain English instead of manually building complex JSON schemas.
• One-click generation — click the new schema generation icon in the schema builder and provide a text prompt to instantly generate a valid JSON schema.
• Automatic validation — generated schemas are validated and can be edited using the visual schema builder or kept as raw JSON.
• Smart parsing — complex schemas with nested objects and arrays are automatically detected and handled appropriately.

​

How to Use

1. Navigate to the Configuration page in your extraction workflow
2. In the JSON Options schema builder section, click the new Generate Schema from Prompt icon (layers icon) next to the Paste Schema button
3. Enter a description of your desired schema, for example:

   “I want to extract invoice information including invoice number, date, total amount, customer name, and a list of line items with product name, quantity, and price.”

4. Click Generate Schema and the AI will create a structured JSON schema matching your requirements
5. The generated schema automatically populates the schema builder where you can review, edit, or use it directly

​

Example Input

Extract purchase order data with PO number, vendor name,
order date, delivery address, and line items containing
item description, quantity, unit price, and total price

​

Generated Schema

{
  "type": "object",
  "properties": {
    "po_number": {
      "type": "string",
      "description": "Purchase order number"
    },
    "vendor_name": {
      "type": "string",
      "description": "Name of the vendor"
    },
    "order_date": {
      "type": "string",
      "format": "date",
      "description": "Date the order was placed"
    },
    "delivery_address": {
      "type": "string",
      "description": "Delivery address"
    },
    "line_items": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "item_description": {
            "type": "string",
            "description": "Description of the item"
          },
          "quantity": {
            "type": "number",
            "description": "Quantity ordered"
          },
          "unit_price": {
            "type": "number",
            "description": "Price per unit"
          },
          "total_price": {
            "type": "number",
            "description": "Total price for this line item"
          }
        },
        "required": [
          "item_description",
          "quantity",
          "unit_price",
          "total_price"
        ],
        "additionalProperties": false
      },
      "description": "List of line items in the purchase order"
    }
  },
  "required": [
    "po_number",
    "vendor_name",
    "order_date",
    "delivery_address",
    "line_items"
  ],
  "additionalProperties": false
}

​

Webhooks

• Complete data delivery — the webhook POST payload contains the same result structure as the GET /v1/extract/results/{record_id} API, including all requested formats (markdown, json, csv, html) with metadata.
• New Webhooks page — configure and manage your webhook URL from the dedicated Webhooks page in the sidebar.
• Request context included — the payload includes filename, output_format, pages_processed, processing_time, created_at, and the original request_config.

​

Example Webhook Payload

{
  "event": "file.processed",
  "success": true,
  "record_id": "12345",
  "status": "completed",
  "result": {
    "markdown": {
      "content": "# Invoice\n\n| Item | Amount |\n|---|---|\n| Widget | $50.00 |",
      "metadata": {}
    }
  },
  "processing_time": 2.45,
  "filename": "invoice.pdf",
  "output_format": "markdown",
  "pages_processed": 3,
  "created_at": "2026-02-20T10:30:00.000Z"
}

​

Document Classification API

• New /api/v1/classify/sync endpoint — classify a single document and get category, document type, and explanation in one response.
• New /api/v1/classify/batch endpoint — classify up to 50 files per request for higher-throughput document triage.
• Routing-ready output — classification labels make it easier to route documents to downstream extraction or review flows.

​

Bounding Boxes in JSON Responses

• Per-field bounding boxes — each extracted field in the JSON output includes a bounding box mapping back to its location in the source document.
• Multi-page support — page-aware resolution of bounding boxes.

​

Example Request

curl -X POST https://api.nanonets.com/api/v1/extract/sync \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@invoice.pdf" \
  -F "output_format=json" \
  -F 'json_options=["invoice_number","total","vendor"]' \
  -F "include_metadata=confidence_score,bounding_boxes"

​

Example Response

{
  "result": {
    "json": {
      "content": {
        "invoice_number": "INV-2024-001",
        "total": 1250.00,
        "vendor": "Acme Corp"
      },
      "metadata": {
        "confidence_score": { "invoice_number": 98, "total": 95, "vendor": 92 },
        "bounding_boxes": {
          "invoice_number": { "page": 1, "x": 0.62, "y": 0.08, "width": 0.15, "height": 0.02, "confidence": 0.99 },
          "total": { "page": 1, "x": 0.70, "y": 0.85, "width": 0.12, "height": 0.02, "confidence": 0.98 },
          "vendor": { "page": 1, "x": 0.10, "y": 0.12, "width": 0.18, "height": 0.02, "confidence": 0.97 }
        }
      }
    }
  }
}

​

Word-Level Bounding Boxes

• Block-level (include_metadata=bounding_boxes) — Existing behavior. Returns one bounding box per paragraph or region detected by layout analysis. Ideal for highlighting sections, table rows, or paragraphs.
• Word-level (include_metadata=bounding_boxes_word) — New. Returns one bounding box per individual word using advanced OCR. Enables fine-grained annotation, search-hit highlighting, and word-by-word document overlays.

​

Response Structure

{
  "result": {
    "markdown": {
      "content": "# Invoice\n\n| Date | Amount |\n| --- | --- |\n| 12/24/18 | $44.78 |",
      "metadata": {
        "bounding_boxes": {
          "success": true,
          "bounding_box_type": "word",
          "elements": [
            {
              "id": "word_0",
              "content": "Invoice",
              "type": "word",
              "page": 1,
              "markdown_line": 0,
              "word_offset": 0,
              "bounding_box": {
                "x": 0.695,
                "y": 0.055,
                "width": 0.051,
                "height": 0.007,
                "confidence": 0.99,
                "page": 1,
                "normalized": true
              }
            }
          ],
          "original_markdown": "# Invoice\n\n| Date | Amount |\n| --- | --- |\n| 12/24/18 | $44.78 |",
          "page_dimensions": { "pages": [{ "page": 1, "width": 1604, "height": 2048 }], "total_pages": 1 }
        }
      }
    }
  }
}

​

API Playground UI

• Output format selection - Choose from Markdown, JSON, CSV/Excel, or HTML output
• Schema Builder - Visual JSON schema editor with support for nested objects, arrays, and enums up to 10 levels deep
• Field List mode - Quick extraction with simple field name arrays
• Metadata options - Enable confidence scores and bounding boxes per field

​

Streaming Extraction Endpoint

• Streaming mode - Content delivered in small chunks as it’s generated
• Batch mode - Content sent all at once when extraction completes

​

v1 Extraction API

• Sync extraction (/v1/extract/sync) - Process documents synchronously with immediate results
• Async extraction (/v1/extract/async) - Queue documents for background processing
• Batch extraction (/v1/extract/batch) - Process up to 50 files in a single request
• Fetch result (/v1/extract/results/<record_id>) - Fetch the results for a single record_id
• Results endpoints (/v1/extract/results) - List and retrieve extraction results with pagination

​

Multi-Page JSON Extraction with Confidence Scoring

​

Bounding Box Extraction API

​

Response Dimensions

​

Streaming & Partial Results

• New streaming extraction endpoint at /v1/extract/stream
• Partial results API to retrieve in-progress extractions
• Improved delimiter handling for chunked responses

​

Billing & Usage APIs

• Credit usage reporting integration with Stripe
• Subscription status tracking
• Document processing limits per plan

​

IP Blocking & Rate Limiting

​

On-Premise License APIs

​

Repetition Detection & Retry

​

Excel & DOCX Processing

​

OpenAI-Compatible Chat Completions API

• PDF and document uploads directly in requests
• All major file types (PDF, Excel, Word, images)
• Drop-in replacement for OpenAI SDK integrations

​

Hierarchy Extraction API

​

Custom Prompt Instructions

​

Expanded File Type Support

• PDF documents
• Excel spreadsheets (.xlsx, .xls)
• Word documents (.docx)
• All major image formats

​

OpenAI SDK Upgrade