Core Concepts

Documents and Extractions

Documents

A document is any file you upload to Documind for processing. Supported formats include:

PDF files
Microsoft Word documents (.docx)
Images (JPEG, PNG, TIFF)

Each uploaded document receives a unique document_id that you use for all subsequent operations.

// Upload response
["123e4567-e89b-12d3-a456-426614174000"]

Extractions

An extraction is a single processing job that extracts structured data from a document using a specific schema. Each extraction:

Belongs to one document
Uses one schema
Can be in different states: pending, processing, completed, failed
May require review depending on confidence scores

You can perform multiple extractions on the same document using different schemas.

Schemas

A schema defines what data to extract from documents. Schemas use a JSON Schema-like format with a special named_entities field for the data you want.

Basic Schema Structure

{
  "type": "object",
  "named_entities": {
    "field_name": {
      "type": "string",
      "description": "What this field represents"
    }
  },
  "required": ["field_name"]
}

Field Types

Documind supports all standard JSON Schema types:

String
Number
Boolean
Array
Object

{
  "invoice_number": {
    "type": "string",
    "description": "The invoice number"
  }
}

{
  "total_amount": {
    "type": "number",
    "description": "Total invoice amount"
  }
}

{
  "is_paid": {
    "type": "boolean",
    "description": "Whether the invoice has been paid"
  }
}

{
  "line_items": {
    "type": "array",
    "description": "Invoice line items",
    "items": {
      "type": "object",
      "named_entities": {
        "description": {"type": "string"},
        "quantity": {"type": "number"},
        "price": {"type": "number"}
      }
    }
  }
}

{
  "vendor": {
    "type": "object",
    "description": "Vendor information",
    "named_entities": {
      "name": {"type": "string"},
      "address": {"type": "string"},
      "tax_id": {"type": "string"}
    }
  }
}

Required Fields

Mark critical fields as required to ensure they’re extracted and flagged for review if confidence is low:

{
  "type": "object",
  "named_entities": {
    "invoice_number": {"type": "string"},
    "total_amount": {"type": "number"},
    "optional_notes": {"type": "string"}
  },
  "required": ["invoice_number", "total_amount"]
}

Extraction Modes

Documind offers three extraction modes with different trade-offs:

Basic Mode (2-6 credits/page)

Single-model extraction for simple documents:

{
  "schema": {...},
  "model": "google-gemini-2.0-flash"  // 2 credits/page
  // or "openai-gpt-4.1"             // 4 credits/page
  // or "openai-gpt-4o"              // 6 credits/page
}

When to use:

Simple, well-formatted documents
Cost is a priority
Speed is important
Review workflow not needed

Limitations:

No confidence scores
No automatic review flagging
Single model may miss edge cases

VLM Mode (10 credits/page)

Vision Language Model-based extraction for image-heavy documents:

{
  "schema": {...},
  "extraction_mode": "vlm"
}

When to use:

Scanned documents
Images with text
Poor quality PDFs
Documents where layout is important

Features:

Uses native image processing
Better for visual documents
No confidence scores

Advanced Mode (15 credits/page)

Multi-model ensemble with confidence scoring:

{
  "schema": {...},
  "review_threshold": 80  // Don't specify model or extraction_mode
}

When to use:

Complex documents
High accuracy required
Review workflow desired
Structured forms and tables

Features:

Multiple models consensus
Confidence scores for every field
Automatic review flagging
Best accuracy

Confidence Scores

Advanced mode provides confidence scores for each extracted field, helping you understand extraction reliability.

Score Calculation

Confidence scores (0-100) are calculated from:

Lexical similarity (40%): How consistent the text is across models
Semantic similarity (60%): How similar the meaning is across models

{
  "results": {
    "invoice_number": "INV-2024-001",
    "total_amount": 1250.00
  },
  "needs_review_metadata": {
    "confidence_scores": {
      "invoice_number": 95.5,
      "total_amount": 78.2
    }
  }
}

Nested Scores

For arrays and objects, scores are nested to match the data structure:

{
  "confidence_scores": {
    "vendor": {
      "name": 92.3,
      "address": 85.1
    },
    "line_items": {
      "0": {
        "description": 88.5,
        "amount": 91.0
      },
      "1": {
        "description": 76.5,
        "amount": 82.0
      }
    }
  }
}

Review Workflow

When required fields have low confidence, extractions are automatically flagged for human review.

Review Threshold

The review_threshold parameter (default: 80) determines when review is needed:

{
  "schema": {...},
  "review_threshold": 85  // Flag fields with confidence < 85%
}

Review Flags

The needs_review_metadata contains flags matching your data structure:

{
  "needs_review": true,
  "needs_review_metadata": {
    "confidence_scores": {
      "invoice_number": 95.5,
      "total_amount": 72.0  // Below threshold!
    },
    "review_flags": {
      "invoice_number": false,
      "total_amount": true    // Needs review
    }
  }
}

Review States

An extraction goes through these states:

Initial Extraction
↓
needs_review = true?
│
├─ No  → Use results immediately
│
└─ Yes → Human reviews
         ↓
         PUT /review/{document_id}
         ↓
         is_reviewed = true
         ↓
         Use reviewed_results

Polling for Review

Poll the extractions endpoint to check review status:

import time

while True:
    response = requests.get(
        f"{BASE_URL}/data/extractions?document_id={document_id}",
        headers=headers
    )
    extraction = response.json()["items"][0]
    
    if extraction["is_reviewed"]:
        results = extraction["reviewed_results"]
        break
    
    time.sleep(10)  # Wait 10 seconds before checking again

Credits System

Documind uses a credit-based pricing model:

Credit Costs

Extraction Mode	Cost per Page
Basic (Gemini 2.0 Flash)	2 credits
Basic (GPT-4.1)	4 credits
Basic (GPT-4o)	6 credits
VLM	10 credits
Advanced	15 credits

Credit Tracking

Monitor your credits via the API:

curl https://api.documind.cloud/api/v1/usage/credits \
  -H "X-API-Key: your_api_key_here"

Response:

{
  "available_credits": 850,
  "total_credits": 1000,
  "lifetime_credits": 5000,
  "subscription_tier": "Professional"
}

Insufficient Credits

When you run out of credits, API calls return 402 Payment Required:

{
  "detail": "Insufficient credits. Please upgrade your plan or wait for your daily credits to refresh."
}

Authentication

All API requests require authentication using API keys passed in the X-API-Key header:

curl https://api.documind.cloud/api/v1/upload \
  -H "X-API-Key: your_api_key_here" \
  -F "files=@document.pdf"

API Key Scopes

API keys can have different permission scopes:

read:extractions - Read extraction results
write:extractions - Create and update extractions
read:api_keys - List API keys
write:api_keys - Create and manage API keys
read:usage - View usage and credits
admin - Full access (admin only)

Organization Keys

API keys can be user-specific or organization-wide, allowing team members to share access.

Error Handling

Documind uses standard HTTP status codes:

Code	Meaning	Action
200	Success	Process the response
400	Bad Request	Check your request parameters
401	Unauthorized	Verify your API key
402	Payment Required	Add credits or upgrade plan
403	Forbidden	Check API key permissions
404	Not Found	Verify document/extraction ID
500	Server Error	Retry or contact support

See the Error Handling Guide for detailed strategies.

Next Steps

Schema Design

Learn best practices for creating effective schemas

Prompt Design

Optimize extraction prompts for better results

Invoice Tutorial

Process invoices end-to-end

API Reference

Explore all API endpoints

Getting Started

Use-Case Tutorials

Advanced Guides

Documents and Extractions

Documents

Extractions

Schemas

Basic Schema Structure

Field Types

Required Fields

Extraction Modes

Basic Mode (2-6 credits/page)

VLM Mode (10 credits/page)

Advanced Mode (15 credits/page)

Confidence Scores

Score Calculation

Nested Scores

Review Workflow

Review Threshold

Review Flags

Review States

Polling for Review

Credits System

Credit Costs

Credit Tracking

Insufficient Credits

Authentication

API Key Scopes

Organization Keys

Error Handling

Next Steps

Schema Design

Prompt Design

Invoice Tutorial

API Reference

Getting Started

Use-Case Tutorials

Advanced Guides

​Documents and Extractions

​Documents

​Extractions

​Schemas

​Basic Schema Structure

​Field Types

​Required Fields

​Extraction Modes

​Basic Mode (2-6 credits/page)

​VLM Mode (10 credits/page)

​Advanced Mode (15 credits/page)

​Confidence Scores

​Score Calculation

​Nested Scores

​Review Workflow

​Review Threshold

​Review Flags

​Review States

​Polling for Review

​Credits System

​Credit Costs

​Credit Tracking

​Insufficient Credits

​Authentication

​API Key Scopes

​Organization Keys

​Error Handling

​Next Steps

Schema Design

Prompt Design

Invoice Tutorial

API Reference

Documents and Extractions

Documents

Extractions

Schemas

Basic Schema Structure

Field Types

Required Fields

Extraction Modes

Basic Mode (2-6 credits/page)

VLM Mode (10 credits/page)

Advanced Mode (15 credits/page)

Confidence Scores

Score Calculation

Nested Scores

Review Workflow

Review Threshold

Review Flags

Review States

Polling for Review

Credits System

Credit Costs

Credit Tracking

Insufficient Credits

Authentication

API Key Scopes

Organization Keys

Error Handling

Next Steps