At Majorbird, we’ve gone beyond the hype and built real AI agents integrated with Odoo — supporting both customers and internal teams. From live chat to PDF-to-order processing, we use a combination of LLMs, workflow automation, and Odoo customization to create intelligent assistants that take action, not just answer questions. Here's how we implemented it recently for two of our customers.
Introduction: AI in the ERP — More Than Just a Chatbot
For many businesses, AI chatbots remain limited to canned responses and simple FAQs. But recently, we aimed higher. By combining Odoo with automation flows (n8n) and external LLMs (Gemini, Claude, etc.), we created an AI system that understands, remembers, reasons — and acts with your ERP data.
The result is a fully integrated assistant that lives inside Odoo but can connect to the outside world, support customers, assist sales teams, and automate follow-ups, all in real time.
From Chatbox to Intelligent Agent: Live Support for Komacut
The most visible application is on www.komacut.com, where the helpdesk chatbox is powered by Odoo and AI. Even though the website is not built in Odoo, the assistant handles incoming conversations and enriches them inside the ERP.
When a customer opens the chat, the system routes the conversation to a salesperson in the closest timezone, attempts to respond based on company knowledge, and generates a full summary once the chat ends. If appropriate, it will create a follow-up task or a new Q&A entry for future reference.
All data — messages, user info, tasks, and Q&As — are stored directly in Odoo, making it a true assistant, not just a chat wrapper.
Personality, Prompt and Parameters: Defining the Agent
Every AI assistant starts with a well-defined personality. In our setup, we created a custom Odoo model to define agent prompts — the set of instructions that guide how the LLM behaves.
These prompts specify the agent’s tone, capabilities, and rules to follow. We also attach LLM parameters such as temperature (randomness), token limits (response length), and preferred models (GPT-4, Claude, Gemini, etc.).
This configuration is fully managed inside Odoo.
The agent can access tools, call other agents, or execute server actions — giving it the ability to make decisions and interact with business data.
Teaching the Bot: Company Knowledge and Memory
To make the AI truly useful, we needed to teach it about the business — products, processes, and terminology. This is done using a combination of static and dynamic memory.
Knowledge is stored in custom Odoo models as text, gathered in three ways:
- Manually written entries
- Automatically generated CSV-like summaries from other Odoo tables
- Extracted content from websites (via HTTP + AI parsing)
Each entry is tagged for context and can be retrieved during conversations. A Q&A model allows AI to propose entries that sales teams can validate and promote into official knowledge.
All of this is indexed into a vector database — a format optimized for fast, semantic search. This makes it possible to implement Retrieval Augmented Generation (RAG): when a user asks a question, the system searches the memory base for related items and sends them along with the prompt and chat context to the LLM.
Memory views in Odoo:
Giving the AI Real Powers: Tools and Actions
An agent that can only talk isn’t enough — we want it to act.
By exposing selected server actions in Odoo and describing them as "tools", we allow the LLM to decide when and how to trigger them. For example, the agent can:
- Fetch a customer’s order history
- Create an opportunity or internal task
- Send a follow-up email
- Even call another agent for specialized tasks
Each tool has parameters, validations, and optional fallback flows. All logic and history is stored, so humans can track or override what the agent did.
Architecture: Odoo + n8n + OpenRouter
To orchestrate all this, we use a layered but lightweight stack:
- Odoo: Stores conversations, knowledge, prompts, tools, user profiles, tasks
- n8n: A free automation engine that manages workflows — such as retrieving memory, calling the LLM, executing server actions, and logging results
- OpenRouter: Connects us to external LLMs — Gemini, Claude, GPT, and more — with flexibility to switch models as needed
This setup is scalable, vendor-agnostic, and low-cost — ideal for companies that want to control their stack.
What We’ve Already Built
Here are some concrete applications already live:
- Live support on the public KOM website: dynamic responses, task creation, knowledge enrichment
- Logged-in support for KOM customers: access to order and product history
- Sales order creation from customer-submitted PDFs
- Internal dashboards + Q&A support (e.g. ALN dashboards through Metabase)
- Internal deployment and company knowledge assistants used by our own team
Conclusion: The Future of ERP Is Intelligent
We believe your ERP should be more than just a system of record — it should think, assist, and take action. By combining Odoo’s flexibility, automation tools, and LLMs, we’ve created a new class of assistant that turns conversations into workflows and knowledge into real business value.
If you're curious how this could apply to your business, we’re ready to show you.
Get a Demo or Build Your Own AI Agent
We're actively building this stack with clients in manufacturing, trading, and services. Want to see what it can do?