By Timothy Jacqmin, co-founder of Nexuro Digital · October 2026
TL;DR
- In Odoo 20, AI is no longer limited to writing text: agents create and update records, work from a file and automate a process described in one sentence.
- Agents can be called by Odoo's automated and scheduled actions, and every run remains viewable so you can correct the behaviour.
- Odoo 20 connects to external AI tools through MCP (Model Context Protocol): an assistant such as ChatGPT or Claude can read and write in your database while keeping existing access rights intact.
- Every AI feature consumes IAP credits, bought on top of licences.
- Our rule: automate an already clean process first, keep human validation at the start, and track consumption from the first month.
"This year our teams put AI in the driver's seat." That is how the official Odoo 20 presentation video opens. After attending the launch at Odoo Experience 2026, we went through the release notes line by line to separate what is genuinely new from what belongs to the demo. Here is what Odoo 20 AI agents can do, how to use them in an SME and what to watch out for.
What is an AI agent in Odoo 20?
An AI agent in Odoo 20 is an assistant built into the database that understands a request in plain language and acts directly in Odoo: it creates, updates and searches records, analyses files and runs automations. It no longer just suggests text: it performs operations, within the limits of the user's access rights.
Each agent is defined by skills (formerly called "topics"): sets of instructions and tools that frame its behaviour in a specific context. Agents get an avatar based on their role, show live what they are doing instead of a plain "AI is thinking", and can ask questions or request permission through clickable buttons.
What Odoo 20 AI agents can do
The official Odoo 20 release notes list five new capabilities: creating records, updating them, working from files, being triggered by automations and generating images.
| Capability | Concrete example |
|---|---|
| Create records | Create a quotation from an order PDF sent by a client |
| Update records | Correct one piece of information across a list of records |
| Query a file | Ask a question about a contract while previewing it |
| Run inside an automation | Process every incoming email according to a rule written in plain English |
| Generate images | Produce a visual or enriched button for a web page or mailing |
| Filter by period | Build a "per quarter" or "per week" view on request |
| Voice dictation | Speak a request instead of typing it |
Two practical details matter day to day: conversations with agents are stored for 30 days after the last message, and if an agent reaches its maximum number of tool calls, it asks for confirmation before continuing. That is a useful safeguard against costly loops.
How do you automate a process in one sentence?
In Odoo 20, you describe the process you want to the agent; it states what it is going to set up, then the automation runs on its own for every new event. This is the central example of the official demo: an email containing an order arrives, the agent creates the sales order, sends the client a confirmation and flags the order to the manager for review.
The release notes describe the mechanics: agents can be called by automated and scheduled actions, they can update themselves to better serve their use case, and the history of automation chats shows how the agent handled each case so you can ask for corrections based on the errors it faced.
The use cases cited by Odoo give a good idea of the potential:
- Auto-assign leads to the right salesperson.
- Create a project with its tasks as soon as a contract is signed.
- Get alerted the moment stock drops below a threshold.
- Identify vendors to renegotiate with from purchase history.
Our field view: the "one sentence" claim is true for setup, not for design. An agent can only correctly automate a process you can describe without ambiguity. Before writing the sentence, ask yourself three questions: who validates, what happens with exceptions, and how will you know the agent got it wrong? That is exactly why Odoo's demo keeps a "flag it for me to review" step.
What is the Odoo 20 MCP connection?
MCP (Model Context Protocol) is a standard protocol that lets an external AI tool connect to an application; Odoo 20 implements it so that ChatGPT, Claude or another assistant can read and write in your Odoo database. Odoo's accounting demo sums it up: you connect your database once, and the external tool accesses the data securely, keeping all existing access rights fully intact.
For an SME, the benefit is twofold:
- Use the assistant your teams already know, rather than imposing a new interface.
- Combine Odoo with other sources in the same conversation (a shared spreadsheet, a mailbox, a business tool).
The flip side is just as clear: an external tool connected with write access can change your data. Odoo's access rights are the main protection, which assumes rights are properly configured. If all your users have administrator rights "for convenience", the MCP connection inherits that flaw. Our article on access rights in Odoo remains a good starting point, especially since Odoo 20 itself simplifies the access rights model.
Which AI model does Odoo 20 use?
Odoo 20 lets you choose an AI provider, then automatically selects that provider's best model for each task. You therefore do not have to choose between fast models and powerful ones: Odoo handles it. Default prompts can be enriched with context and dynamic values, which lets you adapt tone and business rules without development.
For a Belgian company, the choice of provider is not neutral. Check where data is processed and what your GDPR records say before sending customer or payroll data to an agent. This is a governance matter, not a technical setting.
How much does AI cost in Odoo 20?
Every AI feature in Odoo 20 requires IAP (In-App Purchase) credits, bought separately from user licences. Odoo does not publish a price list in its release notes. The cost is therefore tied to usage: an agent plugged into every incoming email will consume far more than an agent queried occasionally.
Our recommendations to keep the budget under control:
- Start with a single process with high volume and low risk, and measure consumption over one month.
- Track credits from the new "My Subscription" page in the user menu.
- Compare the cost with measured time saved, not estimated: how many minutes per order, per lead, per reminder?
For the overall budget of an Odoo project, see our article Odoo 20: the key new features, which also covers the upgrade question.
Where to start with AI in Odoo 20?
The best first use case is a repetitive, already well-defined process where errors are easy to spot and correct. That is how you build your teams' trust before going further.
| Level | Example | Human validation |
|---|---|---|
| Start | Summarise a client file, answer a question about the figures | Read-only, no risk |
| Assist | Prepare a quotation from an email, to validate before sending | Systematic |
| Automate | Assign leads, create a project after signature | Sample checks |
| Delegate | Process incoming orders end to end | Review of exceptions |
It is the same logic we advocate for AI in marketing: concrete uses, not hype, which details our method. And if your goal is specifically to automate marketing, our guide to marketing automation for SMEs remains relevant with Odoo 20.
FAQ
What is an AI agent in Odoo?
An AI agent in Odoo is a built-in assistant that understands a request in plain language and acts directly in the database: it creates and updates records, analyses files and runs automations. In Odoo 20, its behaviour is defined by skills, meaning sets of instructions and tools specific to a context, and it respects the user's access rights.
Can you connect ChatGPT or Claude to Odoo?
Yes. Odoo 20 supports MCP (Model Context Protocol), a standard protocol that lets an external AI tool connect to the Odoo database. Once connected, the tool can read and write data securely, keeping the user's existing access rights fully intact. Access rights therefore need to be properly configured beforehand.
Is AI in Odoo 20 included in the licence?
No. In Odoo 20, every AI feature requires IAP (In-App Purchase) credits, bought separately from user licences. The cost depends on how much the agents are used, not on the number of users. Consumption can be tracked from the "My Subscription" page in the user menu.
How do you automate a process with AI in Odoo 20?
You describe the process to an AI agent in plain language, for example "whenever an email with an order comes in, create the sales order and send the client a confirmation". The agent states what it is going to set up, then the automation runs for every new event. The history lets you check each run and correct the agent.
Can Odoo 20 AI agents make mistakes?
Yes, like any AI system. Odoo 20 includes several safeguards: the agent states its actions before running them, asks for confirmation if it reaches its tool call limit and keeps the automation history for review. We recommend keeping human validation at the start, then moving to sample checks.
Sources
- Odoo, Odoo 20 Release Notes, AI section, September 2026: https://www.odoo.com/odoo-20-release-notes
- Odoo, Meet Odoo 20's best features (official video): https://www.youtube.com/watch?v=YlmN5QcAdLs
- Model Context Protocol, official documentation: https://modelcontextprotocol.io
