Enofirm-Temp/extra-addons/AI/llm_tool_ocr_mistral

Mistral OCR Tool

Extract text from images and PDFs using Mistral AI vision models. Standalone OCR tool that any LLM assistant can use.

Module Type: 🔌 Extension (OCR Tool)

Architecture

┌───────────────────────────────────────────────────────────────┐
│                    AI Consumers                               │
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────────────┐   │
│  │llm_assistant│  │  llm_letta  │  │   llm_mcp_server    │   │
│  └──────┬──────┘  └──────┬──────┘  └──────────┬──────────┘   │
└─────────┼────────────────┼────────────────────┼──────────────┘
          └────────────────┼────────────────────┘
                           ▼
              ┌───────────────────────────────────────────┐
              │   ★ llm_tool_ocr_mistral (This Module) ★  │
              │      Extract Text from Attachment Tool    │
              │  👁️ Vision OCR │ 📄 PDF │ 🖼️ Images       │
              └─────────────────────┬─────────────────────┘
                                    │
                        ┌───────────┴───────────┐
                        ▼                       ▼
    ┌───────────────────────────┐   ┌───────────────────────────┐
    │         llm_tool          │   │        llm_mistral        │
    │    (Tool Framework)       │   │    (Mistral Provider)     │
    └───────────────────────────┘   └───────────────────────────┘

Installation

What to Install

For OCR tool access:

odoo-bin -d your_db -i llm_tool_ocr_mistral

Auto-Installed Dependencies

• llm (core infrastructure)
• llm_tool (tool framework)
• llm_mistral (Mistral AI provider)

Why Use This Module?

Feature	llm_tool_ocr_mistral
OCR	👁️ Mistral vision models
Formats	📄 PDF, PNG, JPG, WEBP
Standalone	🔧 Tool for any assistant
MCP Ready	🔌 Works with external clients

Common Setups

I want to...	Install
Invoice OCR	llm_assistant_account_invoice (includes this)
Chat + OCR	llm_assistant + llm_openai + llm_tool_ocr_mistral
Claude + OCR	llm_mcp_server + llm_tool_ocr_mistral

What is OCR?

Optical Character Recognition (OCR) converts images and scanned documents into machine-readable text. This module provides an LLM tool that:

1. Reads invoices, receipts, bills, and scanned documents
2. Extracts text from handwritten notes and product labels
3. Parses screenshots, photos, and multi-page PDFs

Unlike traditional OCR, Mistral's vision models understand document structure and context, providing accurate text extraction even from low-quality or handwritten sources.

---

Installation

Prerequisites

• Odoo: 18.0 or higher
• Required Modules: llm, llm_mistral, llm_tool
• Mistral API Key: Get one from Mistral AI Platform

Install Steps

1. Install required Odoo modules:

   odoo-bin -d your_database -i llm,llm_mistral,llm_tool,llm_tool_ocr_mistral
   

2. Configure Mistral Provider:

   • Navigate to LLM → Providers → Mistral
   • Add your Mistral API key
   • Click Sync Models

3. Restart Odoo server

---

Quick Start

1. Verify OCR Tool is Available

After installation, verify the tool is registered:

• Navigate to LLM → Tools
• Look for Extract Text from Attachment (OCR)
• Status should be Active

2. Use OCR Tool

The tool is simple - give it an attachment ID, get back parsed text:

# Example: Extract text from attachment ID 5
thread = env['llm.thread'].create({})
thread.message_post(body="Please parse attachment ID 5")
thread.generate_response()

# Assistant uses extract_text_from_attachment tool
# Returns extracted text from the PDF/image

---

How It Works

Tool Interface

Tool Name: extract_text_from_attachment

Input:

{
  "attachment_id": 5
}

Output:

{
  "text": "Extracted text content from the document...",
  "filename": "invoice.pdf"
}

Usage Pattern

1. LLM assistant needs to extract text from a document
2. Assistant knows the attachment ID (from prompt, user message, or search)
3. Assistant calls extract_text_from_attachment(attachment_id=5)
4. Tool uses Mistral vision model to parse the attachment
5. Returns clean, structured text

---

Usage Examples

Basic OCR

# Create thread and ask to parse attachment
thread = env['llm.thread'].create({})
thread.message_post(body="Parse attachment ID 123 and tell me the vendor name")
thread.generate_response()

Invoice Processing

# Invoice assistant with OCR
thread = env['llm.thread'].create({
    'model': 'account.move',
    'res_id': 456,
})
thread.message_post(body="Extract data from attachment ID 789")
thread.generate_response()

With llm_assistant_account_invoice Module

The invoice assistant module uses this OCR tool automatically:

# Install invoice assistant (includes OCR)
odoo-bin -d your_database -i llm_assistant_account_invoice

The assistant's prompt template tells it about attachments and how to use OCR.

---

Screenshots

1. Configure Mistral Provider

Add API key and sync models

2. OCR Models Available

Mistral vision models ready for parsing

3. Text Extraction

Clean text extracted from invoices and receipts

---

Supported Document Types

• Invoices and Receipts: Vendor, date, amount, line items
• Scanned Documents: Paper archives to searchable text
• Handwritten Notes: Forms, signatures, meeting notes
• Product Labels: Codes, descriptions, barcodes
• Screenshots: Error messages, UI text
• Multi-page PDFs: All pages processed automatically

---

Use with External Clients (MCP)

Install llm_mcp_server to expose this tool to external AI clients:

odoo-bin -d your_database -i llm_mcp_server

Compatible Clients:

• Claude Desktop
• Cursor IDE
• Claude Code CLI
• Any MCP-compatible application

Example from Claude Desktop:

You: "Parse attachment ID 123 from my Odoo database"
Claude: [Uses OCR tool via MCP] "The document contains..."

This lets you extract text from Odoo attachments without opening Odoo UI.

---

Configuration

Tool Consent

Configure in LLM → Tools → Extract Text from Attachment:

• Always Allow (recommended) - Auto-execute
• Ask Permission - User approves each use
• Never Allow - Disable tool

---

Troubleshooting

Tool not found?

• Check llm_tool_ocr_mistral is installed
• Verify tool is active in LLM → Tools

No vision models?

• Go to LLM → Providers → Mistral
• Click Sync Models
• Verify API key is valid

Extraction fails?

• Verify attachment ID exists
• Check attachment is image/PDF format
• Ensure Mistral vision model is available

Poor quality results?

• Use higher resolution images (300+ DPI)
• Ensure text is legible
• Try clearer scan/photo

---

Related Modules

llm_assistant_account_invoice - Invoice data entry assistant using OCR llm_knowledge_mistral - Knowledge base with OCR support llm_mcp_server - Expose tools to external MCP clients

---

Documentation

• Full Documentation: doc/index.rst
• GitHub: https://github.com/apexive/odoo-llm

License

LGPL-3 - https://www.gnu.org/licenses/lgpl-3.0.html

---

© 2025 Apexive Solutions LLC