Practical AI automation in Odoo means letting an AI model read documents and data, prepare a draft or an answer, and then leave the decision to a person with the right Odoo permissions. Every action runs under a real user's access rights, is recorded in an audit log and, for write actions, can be undone. I design and build such automations for companies that already run Odoo, starting with one process that wastes the most hours.
What does useful AI in Odoo look like?
Vendor bills from e-mail and photos
Incoming invoices sent by e-mail or photographed on a phone are read by an AI model and turned into draft vendor bills in Odoo, with the original document attached. An accountant checks and posts them. I use this setup in my own company's bookkeeping.
Questions answered from your data
Managers ask an AI assistant about overdue invoices, open purchase orders, stock by serial number or the sales pipeline and get answers taken from Odoo, within their own permissions.
Routine updates with undo
An assistant prepares a change to records, for example in a project, under the user's own permissions. The user sees what will change before confirming, and a wrong change can be reversed.
Imports from messy lists
Participant lists, product lists or contact lists pasted from e-mails or spreadsheets are structured by AI and shown for review before any record is created.
How do approvals, undo and audit work?
The guardrails are designed in from the start, not added later:
- Per-user keys. Each person connects with their own Odoo API key, so Odoo's normal access rights and record rules apply. Nobody gets more through AI than they have in Odoo.
- Read-only by default. Write tools are enabled only for the processes and users that need them.
- Drafts, not postings. Documents with financial effect are created as drafts. Confirming them is a separate human step.
- Undo. Write actions record what was changed, so a wrong change can be reversed.
- Audit log. Who asked for what, which tool ran, which records were touched and when.
- Isolation. Each client's connection runs as its own separate instance, not as a shared service mixing companies' data.
Where does MCP fit in?
The Model Context Protocol is an open standard for connecting AI assistants to external tools. An MCP connection to Odoo exposes a defined set of tools, such as search partners, list overdue invoices or update a task, instead of giving an assistant free access to the database. This keeps behaviour predictable and makes the audit log meaningful. Under the hood the tools call Odoo's External API with the user's own key.
How is data protection handled?
This page is informational, not legal advice; for your specific case, involve your data protection officer or lawyer.
- Map the data first. We list which Odoo models and fields the automation reads. Payroll, health data and other special categories are excluded unless there is a clear legal basis.
- Choose the provider deliberately. Use business or API terms where the provider contractually does not train models on your data, and check where the data is processed.
- Contracts. Where a provider processes personal data for you, a data processing agreement under Article 28 of the GDPR is needed.
- Minimise. Send the model only what the task needs: a document for extraction, not the whole partner history.
- Keep the trail. The audit log and the human approval step make it possible to explain later what happened and why.
How do we start?
Pick one process
The one where people spend the most time on repetitive reading or typing, and where a mistake is caught by a human check anyway.
Check access rights
AI inherits your permission model. If groups and record rules are messy, that is fixed first, usually as part of a health check.
Pilot with a few users
Run the automation on real work for a few weeks with two or three users, measure how many drafts needed correction, and adjust.
Extend or stop
If the pilot saves time without creating extra checking work, extend it. If it does not, stop it. Both are useful results.
What do I not build?
Autonomous agents that post entries, send payments or e-mail customers without a person confirming. AI features that bypass Odoo's own permissions through an administrator account. Automations whose results nobody can trace back to a source document. If a simple automated action or a custom module solves the problem without AI, I will recommend that instead.
Tell me about your Odoo
Describe your situation in a few lines. We reply by e-mail.
Frequently asked questions
Can an AI assistant post accounting entries on its own?
Technically yes, and I do not recommend it. In the setups I build, AI creates drafts and proposals, and a person with the right Odoo permissions confirms them. Posting, payments and anything that changes balances stays a human decision, with the action recorded in the audit log.
Does our Odoo data leave the company?
The part that the AI model needs to read does, unless the model runs on your own infrastructure. We decide which models and fields are exposed, choose a provider whose business terms exclude training on your data, and sign the data processing agreements before anything is connected.
What is MCP and do we need it?
The Model Context Protocol is an open standard that lets AI assistants call tools in other systems in a controlled way. For Odoo it means an assistant can search records or prepare an update through defined tools instead of free database access. You need it only if your team works in an AI assistant daily.
Which Odoo versions does this work with?
Read access through the External API works on Odoo Online, Odoo.sh and self-hosted databases. Write actions with approval and undo are simplest on recent versions; my current setup runs on Odoo 19 Enterprise. Older versions are possible after a short technical check.