Skip to content

Mistral OCR Integration

This guide covers setting up Mistral AI OCR for extracting text from PDF documents.

Overview

Mistral OCR provides high-quality text extraction from PDFs, including scanned documents. When enabled, the assistant can process PDF documents sent via WhatsApp, Slack, or uploaded directly.

Capabilities: - Extract text from PDFs (including scanned/image-based PDFs) - No character limit on extracted text - Configurable model and API endpoint

Prerequisites

  • Mistral AI account (or any OpenAI-compatible API endpoint)
  • PDF documents to process

Step 1: Get an API Key

  1. Go to https://console.mistral.ai/
  2. Sign up or log in
  3. Go to API Keys
  4. Click Create new key
  5. Copy the API key

Option B: Azure OpenAI (with Mistral model)

If using Azure OpenAI with a Mistral deployment: - Set the Base URL to your Azure endpoint - Use your Azure API key

Option C: Self-hosted Compatible Endpoint

Any OpenAI-compatible API that implements the Mistral OCR endpoint format: - Set the Base URL to your server endpoint - Ensure it supports the /v1/ocr or equivalent endpoint

Step 2: Configure in Settings

  1. Go to Settings > Integrations > Mistral OCR
  2. Enable the Mistral OCR integration
  3. Paste your API key
  4. (Optional) Adjust the Base URL if using Azure or a custom endpoint
  5. (Optional) Set the Model (default: mistral-ocr-latest)
  6. Click Save
  7. Click Test Connection to verify

Or use environment variables:

MISTRAL_OCR_ENABLED=true
MISTRAL_OCR_API_KEY=your-api-key
MISTRAL_OCR_BASE_URL=           # Leave empty for default Mistral AI
MISTRAL_OCR_MODEL=mistral-ocr-latest

How It Works

Automatic PDF Processing

When a PDF is sent via WhatsApp or Slack, the assistant automatically:

  1. Downloads and processes the PDF using Mistral OCR
  2. Extracts the text content
  3. Returns the extracted text to you for further use (e.g., summarization, analysis)

You can then ask the assistant to save the result to Notion or Nextcloud separately.

Example Conversations

You: Can you read this invoice? [attaches invoice.pdf]
Assistant: I've extracted the text from the PDF. Here's what it says:
[full extracted text]

You: Summarize this document [attaches report.pdf]
Assistant: [Extracts text, then summarizes the content]

Manual PDF Processing

You can also ask the assistant to process PDFs explicitly:

You: Extract all text from the PDF I sent earlier
Assistant: [Returns the extracted text]

Configuration Options

Setting Default Description
mistral_ocr.enabled false Enable/disable Mistral OCR
mistral_ocr.api_key (empty) API key. Falls back to main LLM key if empty
mistral_ocr.base_url (empty) API base URL. Leave empty for Mistral AI
mistral_ocr.model mistral-ocr-latest OCR model to use

Compatible Providers

Any OpenAI-compatible API that implements the Mistral OCR endpoint format works:

Provider Base URL Notes
Mistral AI (default) Official Mistral OCR API
Azure OpenAI https://{resource}.openai.azure.com/openai/deployments/{deployment}/ Requires Mistral model deployment
Local server http://localhost:8000/v1 Self-hosted compatible server

Troubleshooting

"Mistral OCR connection failed"

  • Verify the API key is correct
  • Check that you have billing set up on Mistral AI (free tier has limits)
  • If using Azure, verify the endpoint URL is correct

"No API key configured"

  • Set mistral_ocr.api_key in Settings
  • Or ensure your main LLM API key is set (Mistral OCR falls back to it)

PDF text extraction returns empty

  • The PDF may be a scanned image without text
  • Mistral OCR should handle most scanned PDFs, but some low-quality scans may fail
  • Try a higher quality scan if available

Model not found error

  • Verify the model name is correct (mistral-ocr-latest)
  • If using a custom provider, check which models are available

Rate Limits

Mistral AI OCR pricing is based on pages processed. Check current pricing at https://mistral.ai/pricing.

Azure OpenAI uses standard Azure pricing for the deployed model.

Security Considerations

  • API Key: Store securely — it grants access to your Mistral AI quota
  • PDF Content: PDF files are sent to the Mistral API for processing

API References

Last Updated: March 2026