AI in Odoo is no longer a demo. Odoo 19 ships AI agents, AI-computed fields, AI automations and an AI live chat, and Odoo 20 adds AI to Sales, Purchase, Inventory, Project, Helpdesk and Marketing, plus a built-in MCP server so outside assistants like Claude and ChatGPT can work with your Odoo data. Around that sits a whole ecosystem of AI services - language models, document AI, speech, automation platforms and analytics. This guide maps all of it: what Odoo does natively, what you can connect, what it is good for, and how to roll it out without losing control of your data.
Key takeaways
- Built in (Odoo 19 Enterprise): AI agents that answer from your documents with citations, AI text drafts in chatter and email, AI fields and AI server actions, AI live chat that creates CRM leads, AI document sorting and call transcription.
- Odoo 20 spreads AI agents across Sales, Purchase, Inventory, Project, Helpdesk, Calendar, Marketing and Accounting reports, and adds an MCP server at
/mcpfor external AI assistants. - Models: Odoo 19's AI runs on OpenAI (GPT-4o, GPT-4.1, GPT-5 and mini variants) or Google Gemini (2.5 Pro and Flash). Self-hosted databases use your own API key.
- Outside Odoo: Claude, local open-source models, document AI, automation platforms like n8n, Make and Zapier, and BI tools all connect through Odoo's APIs or MCP.
- Start with one or two measurable use cases, keep a human approving anything that posts, sends or pays, and respect Odoo's access rights.
AI built into Odoo 19
Odoo 19 introduced a proper AI framework: an AI app with agents, plus AI features inside other apps. These are Enterprise features.
| Feature | What it does | Where |
|---|---|---|
| AI agents | Chat assistants you configure with instructions, a response style, a model, and sources - uploaded files, web pages or Documents - indexed so answers come with citations. Can be limited to answer only from those sources. | AI app |
| AI text drafts | Summarise the chatter, write a follow-up, rewrite selected text, recap a call - from the email composer and the chatter. | All apps with chatter, Knowledge, Accounting |
| Ask AI on your data | Natural-language questions such as "top 5 sales reps by revenue" or "products with less than 5 in stock", answered from your records. | AI app |
| AI fields | A field whose value is computed by an AI prompt from the record - a sentiment, a category, a short summary. Can be created in Studio. | Studio, any model |
| AI server actions | Automations that update record fields with AI, for example classify every new ticket or lead. | Automation rules |
| AI live chat | An AI agent answers website visitors in Live Chat and can create CRM leads from the conversation. | Website, Live Chat, CRM |
| AI document sorting | Documents dropped into a folder are sorted into sub-folders following a prompt you write. | Documents |
| Call transcription | Calls in the Phone app are transcribed and summarised. | Phone (VoIP) |
| Recruitment and Sign | AI assistance for applicants and for documents to sign. | Recruitment, Sign |
| Document digitization | Reads vendor bills, customer invoices, expense receipts and bank statements and fills the record. Uses Odoo's in-app purchase (IAP) credits. | Accounting, Expenses |
| Lead enrichment | Adds company information to leads from the email domain, also an IAP service. | CRM |
Which AI models Odoo 19 uses
Each agent picks a model from two providers: OpenAI (GPT-4o, GPT-4.1, GPT-4.1 mini, GPT-5, GPT-5 mini and older GPT-4 / GPT-3.5) or Google (Gemini 2.5 Pro and Gemini 2.5 Flash). Agent sources are turned into embeddings with OpenAI or Google embedding models and stored in your PostgreSQL database. On a self-hosted Odoo 19 database you enter your own OpenAI or Google AI API key in Settings, and the provider bills you for usage. Check how AI usage is handled for your plan on Odoo Online and Odoo.sh - our hosting comparison explains the three options.
What Odoo 20 adds
Odoo 20 takes AI from a separate app to almost every app. In the Odoo 20 source there are dedicated AI modules for:
- Sales, Purchase and Inventory - agents that work with quotations, purchase orders, stock and margins.
- Project, Timesheets, Helpdesk and Calendar - agents for tasks, time, tickets and scheduling.
- Email Marketing, Marketing Automation and Social Marketing - AI-written posts and campaigns, including campaigns created from live chat.
- Accounting reports - AI-assisted report analysis and audit review.
- Website builder and eCommerce - AI content in the website editor and shop, AI live chat on blog, events, eLearning, appointments and shop pages.
- Agent skills and tools - reusable skills that either guide an agent or let it run approved actions.
- An MCP server - see below.
Our Odoo 20 release notes cover the rest of the release.
The Odoo 20 MCP server: connect Claude, ChatGPT and other assistants
The Model Context Protocol (MCP) is an open standard that lets AI assistants call tools in other systems. Odoo 20's AI MCP Server module turns your database into an MCP server:
- The server URL is your Odoo URL plus
/mcp, for examplehttps://yourcompany.odoo.com/mcp. - Each user generates an API key with the MCP scope (user avatar > Security tab), or the client signs in with OAuth.
- Add it to an MCP client - Claude, ChatGPT, an IDE or your own agent - as a credential, never in a plain text field.
The assistant can then use the Odoo tools you have enabled for AI, with that user's access rights. Ask "which customers have overdue invoices over 30 days?" or "draft a purchase order for the items below minimum stock" from your AI assistant, and the work happens in Odoo.
On Odoo 17 to 19 the same idea works through community or custom MCP servers that talk to Odoo's external API (XML-RPC, JSON-RPC, and the newer JSON-2 API in Odoo 19). We build and host these for clients, scoped to read-only or to specific actions.
Every AI service you can use with Odoo
| Service type | Examples | How it connects to Odoo | Good for |
|---|---|---|---|
| Language models (cloud) | OpenAI GPT, Google Gemini | Native in Odoo 19 and 20 AI | Agents, drafts, AI fields, automations |
| AI assistants | Claude, ChatGPT, other MCP clients | Odoo 20 MCP server, or a custom MCP server for older versions | Asking questions and acting on Odoo from your assistant |
| Other model APIs | Anthropic Claude API, Mistral, Azure OpenAI | Custom module or middleware calling the provider's API | A preferred provider, data-region requirements |
| Private, self-hosted models | Open-weight models such as Llama, Mistral or Qwen served with Ollama or vLLM | Custom module using an OpenAI-compatible API on your own server | Sensitive data that must not leave your infrastructure |
| Document AI / OCR | Odoo digitization, Google Document AI, Azure AI Document Intelligence, AWS Textract | Native for invoices, bills, receipts and bank statements; custom for other documents | Delivery notes, certificates, contracts, customs forms |
| Speech | Odoo Phone transcription, speech-to-text APIs | Native in Odoo Phone; custom for other call systems | Call summaries, logging calls to CRM and Helpdesk |
| Automation platforms | n8n, Make, Zapier | Their Odoo connectors plus AI steps, over Odoo's API | Email-to-order, ticket triage, cross-app workflows without code |
| Analytics and BI | Metabase, Power BI, Looker Studio | Read-only database or API connection | Dashboards across companies and databases, AI on top of clean data |
| Website chat and search | Odoo AI live chat, custom "Ask AI" assistants | Native in Odoo, or a custom module | Answering visitors 24/7 and capturing leads |
| AI coding assistants | Claude Code, GitHub Copilot, Cursor | Used by developers on Odoo modules | Faster custom modules, upgrades and tests |
For analytics specifically, see Odoo + Metabase; for online stores, Odoo with Shopify and WooCommerce.
AI use cases by department
| Team | Use cases |
|---|---|
| Sales and CRM | Lead scoring and classification with AI fields, follow-up email drafts, call summaries logged on the opportunity, quotation drafts from a customer email, AI live chat qualifying website visitors. |
| Accounting | Bill and receipt digitization, bank statement import from PDF, explaining unusual journal entries, report review before closing, collection email drafts. |
| Purchase and Inventory | Reorder suggestions, purchase orders from supplier quotes, product descriptions and categories, questions like "what moves most between these two warehouses?". |
| Customer service | Ticket triage and tagging, answer suggestions from your knowledge base with citations, customer sentiment, summaries for escalation. |
| HR and recruitment | Job descriptions, applicant summaries, internal policy assistant answering from your handbook. |
| Marketing and website | Blog and product copy, social posts, email campaigns, SEO meta descriptions, translations. |
| Management | Plain-language questions on sales, stock and cash; daily briefings; dashboards across several companies or databases. |
Privacy, security and cost
- Data leaves your server when you use a cloud model. Review the provider's data-use terms and region; use a self-hosted model for data that must stay in-house.
- Access rights still apply. MCP and API keys act as a user, so create dedicated users with only the rights an assistant needs, and prefer read-only to start.
- Keep a human in the loop for anything that posts, sends, pays or deletes: let AI draft, people approve.
- Public chatbots should only see public information - never let a website agent read customer or financial records.
- Watch usage cost. Pick smaller models (GPT-5 mini, Gemini 2.5 Flash) for bulk tasks like classification, and stronger ones only where quality matters.
- Check results - AI can be confidently wrong. Citations from agent sources and spot checks keep it honest.
A practical rollout plan
- Pick one or two use cases with a clear measure - hours spent entering bills, response time on tickets, leads qualified per week.
- Clean the data first. AI on messy product names and duplicate customers gives messy answers.
- Start with built-in features (digitization, drafts, an agent on your own documents) before custom work.
- Add integrations - MCP, automation platforms, private models - where the built-in features stop.
- Train the team on good prompts and on what AI may and may not do in your company.
- Review after a month: time saved, error rate, cost, and the next use case.
How SNM Digitech helps
- Setting up Odoo AI agents, sources, AI fields and automations around your processes.
- Custom AI modules: private models, other providers, document types Odoo does not read out of the box.
- MCP servers for Odoo 17, 18 and 19, and Odoo 20 MCP setup with safe users and scopes.
- Our SNM AI Workspace, which connects AI assistants, reports and dashboards to one or many Odoo databases.
Frequently asked questions
Does Odoo have built-in AI?
Yes. Odoo 19 Enterprise includes an AI app with agents, AI text drafts, AI fields, AI server actions, AI live chat, AI document sorting and call transcription. Odoo 20 adds AI to many more apps and a built-in MCP server.
Which AI models does Odoo use?
In Odoo 19 each AI agent uses either an OpenAI model (such as GPT-4o, GPT-4.1, GPT-5 or GPT-5 mini) or a Google Gemini model (Gemini 2.5 Pro or Flash).
Can I connect ChatGPT to Odoo?
Yes. Odoo 19's AI features use OpenAI models natively, and in Odoo 20 ChatGPT and other MCP clients can connect to Odoo's MCP server at your Odoo URL plus /mcp, using an API key with the MCP scope or OAuth.
Can I use Claude with Odoo?
Yes. Claude can connect to the Odoo 20 MCP server, or to a custom MCP server for Odoo 17 to 19. A custom module can also call the Claude API directly for features inside Odoo.
Is Odoo AI available in the Community edition?
Odoo's AI app and most AI features are Enterprise. Community users can still add AI through custom modules, automation platforms or an MCP server using Odoo's external API.
Do I need my own OpenAI or Google API key?
On a self-hosted Odoo 19 database, yes - you enter it in Settings and the provider bills you for usage. Check with Odoo how AI is provided for your Odoo Online or Odoo.sh plan.
Can I run AI on my own server without sending data to the cloud?
Yes, with an open-weight model served on your own hardware (for example with Ollama or vLLM) and a custom Odoo module that calls it. Quality depends on the model and hardware you choose.
Is it safe to let AI change records in Odoo?
It can be, with dedicated users, limited access rights, approved tools only, and a person approving anything that posts, sends or pays. Start read-only.
How does Odoo read invoices and receipts?
Odoo's document digitization reads vendor bills, invoices, expense receipts and bank statements and fills in the record automatically. It is an in-app purchase service that uses credits.
See Odoo for yourself
Start a free Odoo trial and click through Sales, Inventory, Accounting and the website builder with your own data. When you want it set up around how your business works, that is where we come in.
Put AI to work in your Odoo
Tell us where your team loses the most time. We will show you what Odoo AI can do there today, and build what it cannot.
Based on the Odoo 19.0 and Odoo 20 source code and module descriptions, reviewed in October 2026. Feature availability can differ by edition, hosting and plan, and AI providers change their models often.