Create Invoice Analysis Thread

1. Manual Way:
   • Go to LLM → Threads → Create
   • Link to your invoice using the record picker
   • Select "Invoice Analysis Assistant"
   • Start chatting!
2. Future Enhancement: Add "Ask AI" button to invoice form

Example Conversations

Basic Analysis:

User: What is the vendor and total amount?
Assistant: The vendor is Acme Corp and the total amount is $5,420.00

Document Parsing:

User: Parse attachment 123
Assistant: [uses llm_mistral_attachment_parser]
Assistant: I extracted: Invoice #INV-2024-001 from Acme Corp
         Date: 2024-01-15, Total: $5,420.00
         Line items:
         - Product A: $2,000.00
         - Product B: $3,420.00

Data Retrieval:

User: Find other invoices from this vendor
Assistant: [uses odoo_record_retriever]
Assistant: Found 3 other invoices from Acme Corp:
         - INV-2023-045: $3,200.00 (Paid)
         - INV-2023-089: $1,850.00 (Paid)
         - INV-2024-002: $4,100.00 (Draft)

Validation:

User: Check if this invoice looks correct
Assistant: I've reviewed the invoice and found:
         ✓ Vendor VAT number present
         ✓ Invoice has line items
         ⚠️ Warning: Invoice date is in the future
         ⚠️ Warning: No payment terms specified

Context Awareness

• Automatically accesses invoice fields via related_record proxy
• No need to specify invoice ID in every query

Available Tools

1. llm_tool_ocr_mistral: Parse PDFs/images with Mistral OCR
2. odoo_record_retriever: Search and retrieve Odoo records
3. odoo_record_updater: Update invoice fields (requires consent)
4. odoo_model_inspector: Inspect model structure

Intelligent Instructions

The assistant knows how to:

• Extract data from OCR results
• Validate invoice consistency
• Handle accounting workflows
• Respect user consent for updates
• Follow best practices for financial data

Pure Configuration Module

This module contains no Python code - just XML data files:

llm_assistant_account_invoice/
├── __manifest__.py              # Dependencies and metadata
├── __init__.py                  # Empty (no code needed)
├── data/
│   ├── llm_prompt_invoice_data.xml  # Invoice-specific prompt template
│   └── llm_assistant_data.xml       # Assistant configuration
├── views/
│   └── account_move_views.xml   # "Process with AI" button on invoice form
└── static/
    └── description/
        └── index.html           # App store description

How It Works

1. Uses existing prompt template from llm_assistant
2. Provides invoice-specific context via default_values
3. References existing tools via XML ref()
4. Leverages llm_thread for record linking

Required Modules

• account: Core Odoo accounting (vendor bills)
• llm_assistant: Base LLM assistant framework
• llm_tool_ocr_mistral: Mistral OCR tool for parsing documents

Transitive Dependencies

These are pulled in automatically:

• llm: Core LLM provider/model system
• llm_thread: Thread management with record linking
• llm_tool: Tool registration and consent
• llm_mistral: Mistral AI provider

1. Mistral Provider Setup

1. Go to Settings → LLM → Providers
2. Find or create "Mistral AI" provider
3. Enter your API key
4. Click "Sync Models"
5. Verify OCR models appear (e.g., "mistral-ocr-latest")

2. Assistant Settings

The assistant is pre-configured with sensible defaults:

• Name: Invoice Analysis Assistant
• Code: invoice_analyzer
• Model: account.move
• Public: Yes (all users can access)
• Default: Yes (default assistant for invoices)

Customize in Settings → LLM → Assistants if needed.

Access Control

• Thread creation: Requires llm_thread permissions
• Tool usage: Controlled by llm_tool consent system
• Invoice access: Controlled by account module groups

Tool Consent

• llm_tool_ocr_mistral: Requires consent (accesses attachments)
• odoo_record_retriever: No consent (read-only)
• odoo_record_updater: Requires consent (modifies data)

Users must approve tool execution before sensitive operations.

Add Custom Instructions

Edit the assistant's default_values in data/llm_assistant_data.xml:

<field name="default_values"><![CDATA[{
    "role": "Invoice Analysis Assistant",
    "goal": "...",
    "instructions": "Your custom instructions here..."
}]]></field>

Add More Tools

Reference additional tools in the tool_ids field:

<field name="tool_ids" eval="[(6, 0, [
    ref('llm_tool_ocr_mistral.llm_tool_ocr_mistral'),
    ref('llm_tool.llm_tool_odoo_record_retriever'),
    ref('your_module.your_custom_tool'),  <!-- Add this -->
])]" />

Create Multiple Assistants

Duplicate the record with different configurations:

• Invoice Validator (read-only, no updater tool)
• Invoice Data Entry (focuses on OCR + updates)
• Invoice Approver (workflow-focused)

For Users

1. Upload documents first, then ask AI to parse them
2. Review extracted data before confirming updates
3. Use specific questions for better results
4. Link threads to invoices for context-aware responses

For Administrators

1. Monitor tool consent: Review which users approve updates
2. Track assistant usage: Check thread activity
3. Customize instructions: Tailor to your accounting workflows
4. Train users: Show examples of effective prompts

"No OCR model found"

Solution: Sync models from Mistral provider settings

"Mistral provider not found"

Solution: Install and configure llm_mistral module with your API key

"Tool requires consent"

Expected: User must approve before tool executes. This is by design for security.

Thread not linked to invoice

Solution: Use record picker to link thread to account.move record