Artificial Intelligence is no longer an add-on in ERP ecosystems—it is becoming the foundation. With the release of Odoo 19, this shift is clearer than ever. Instead of layering AI features on top, Odoo has embedded intelligence deep within its core architecture using Retrieval-Augmented Generation (RAG), vector embeddings, and configurable AI agents.

For technical implementers and developers, this raises critical questions:
How does it actually work? Where does the data live? What can be customised—and what still remains a black box?

In this technical guide, we break down Odoo 19’s AI architecture, explore its internal mechanics, and highlight practical ways to extend it for real business value.

The 3 Pillars of Odoo AI Architecture

Odoo 19’s AI framework is built on three tightly integrated pillars:

1. Vector Database (RAG)
Enables AI to retrieve relevant business data using embeddings before generating responses.

2. AI Agents
Configurable assistants that combine prompts, tools, and contextual data to perform tasks.

3. AI Tools
Server-side actions that allow agents to interact directly with Odoo data and workflows.

Together, these components create a hybrid AI system—one that doesn’t just generate text but actually understands and acts within your ERP environment.

Vector-Based RAG: The Foundation of Contextual Intelligence

At the heart of Odoo’s AI lies RAG (Retrieval-Augmented Generation), a method that enhances AI responses with real, contextual data.

RAG Response = LLM Output + Retrieved Context

Instead of relying purely on pre-trained knowledge, Odoo converts documents into embeddings—numerical representations stored in a vector database.

How It Works

• Documents (PDFs, knowledge articles, etc.) are broken into chunks
• Each chunk is converted into a vector using an embedding model
• These vectors are stored in PostgreSQL using the pgvector extension
• When a query is made, similar vectors are retrieved and passed to the AI

This approach ensures that responses are grounded in your actual business data—not generic assumptions.

Inside the Odoo Vector Store

The vector database is built around the ai.embedding model. Importantly:

• Embeddings are not stored in source records
• All vectors are centralised for efficient querying
• Only content from agent.source is embedded

Data Flow Pipeline

1. Source is added (document, article, etc.)

2. Scheduled jobs trigger processing

3. Content is chunked

4. Embeddings are generated via LLM APIs

5. Vectors are stored in PostgreSQL

Once indexing is complete, the system can perform fast similarity searches across all embedded content.

Key Insight

Despite the advanced search experience, Odoo does not use embeddings for standard models like CRM, products, or contacts.

Instead:

• RAG = document intelligence
• Tools = operational data access

This distinction is critical when designing AI-driven workflows.

Odoo AI Agents: The Decision Layer

AI Agents are the operational brain of the system. Think of them as intelligent assistants that combine:

• System prompts
• Business context
• Retrieved knowledge
• Actionable tools

Prompt Assembly Flow

When a user interacts with an agent, Odoo builds the request in layers:

1. System-level instructions

2. Agent-specific prompts

3. Context (user, date, records)

4. RAG-retrieved data

5. User query

6. Tool execution (if needed)

7. Final response

This structured pipeline ensures that responses are both context-aware and actionable.

Recursive Intelligence

Agents can call tools multiple times before responding.

• Default max recursive calls: 20
• Tool calls per run: 20

While powerful, this also increases token usage—something teams must monitor in production environments.

AI Tools: From Insight to Execution

If RAG provides knowledge and agents provide reasoning, AI tools provide action.

These are essentially server actions exposed to the AI layer.

What AI Tools Can Do

• Retrieve records (CRM, sales, inventory)
• Open views (list, form, pivot)
• Create new records (leads, tasks)
• Execute business logic
• Generate reports

When a user asks: “Show me my open opportunities”—the agent doesn’t search embeddings.
It calls a tool that executes a database query.

This makes the system:

• Faster
• More accurate
• Cost-efficient

Topics: The Bridge Between AI and Data

Topics act as containers that group tools together.

Each topic defines:

• What the agent can access
• Which actions it can perform

Extending Capabilities

You can easily extend Odoo AI by:

1. Creating a server action

2. Linking it to a topic

3. Assigning the topic to an agent

This modular approach allows developers to add powerful capabilities without modifying core code.

Customisation Options: What You Can (and Can’t) Do

Easy, No-Code Adaptations

• Modify system prompts
• Add knowledge sources
• Configure response styles
• Create new topics
• Build simple AI tools

Advanced Customisation

• Custom AI tools with complex logic
• Integration with external AI providers
• Backend parameter tuning

Current Limitations

• No native RAG for standard Odoo models
• Limited support for custom embedding pipelines
• Hardcoded API endpoint structures
• UI-level constraints requiring JavaScript changes

While flexible, the architecture still has boundaries—especially for organisations looking to deeply customise AI behaviour.

Odoo API Endpoints: Flexibility vs Constraints

One of the biggest questions for enterprises is whether they can use their own LLM infrastructure.

Current Reality

• Base URL is configurable
• API endpoints are hardcoded
• Designed specifically for OpenAI and Google APIs

Implication

Using alternatives like:

• AWS Bedrock
• Azure OpenAI
• Vertex AI

May require additional adaptation unless they mimic the expected API structure.

This is a key consideration for enterprises focused on data sovereignty or cost optimisation.

Practical Implications for Implementers

For Business Teams

• Focus on high-quality knowledge content
• Use AI agents for guided workflows and automation
• Understand that RAG ≠ full database search

For Developers

• Ensure PostgreSQL includes pgvector
• Build extensions via server actions (tools)
• Avoid overloading RAG with unnecessary data
• Monitor token usage in recursive workflows

Strategic Insight

The real power of Odoo AI lies in combining:

• Structured data (via tools)
• Unstructured data (via RAG)

Not replacing one with the other.

The Future of AI in Odoo

Odoo 19 lays a strong foundation—but it is just the beginning.

Expected Evolution

• Expansion of RAG to more models
• Faster and cheaper embeddings
• Broader LLM compatibility
• Enhanced built-in tools
• Deeper automation capabilities

High-Potential Use Cases

• Intelligent helpdesk automation
• AI-assisted project management
• Context-aware CRM recommendations
• Automated knowledge generation

As the ecosystem evolves, the boundary between ERP and AI platform will continue to blur.

Final Thoughts

Odoo 19’s AI architecture represents a shift from passive systems to intelligent platforms. By combining RAG, embeddings, and agent-driven workflows, it enables businesses to move beyond static processes toward adaptive, data-driven operations.

However, success depends on understanding the mechanics—what is powered by embeddings, what is driven by tools, and where customisation delivers the most impact.

For organisations willing to invest in the right architecture and content strategy, Odoo AI is not just an upgrade—it’s a transformation layer.

About Aarav Solutions

Aarav Solutions is a global technology consulting and services company specializing in digital transformation for telecom and enterprise businesses. With deep expertise in Odoo, AI, and OSS/BSS platforms, Aarav Solutions helps organizations design, deploy, and scale intelligent systems that drive operational efficiency and customer experience.

Let’s Build Your AI-Powered Odoo Ecosystem

Looking to implement or scale AI within your Odoo environment?

Connect with our experts at cocreate@aaravsolutions.com and explore how we can help you unlock the full potential of AI-driven ERP.