Odoo 19 adds a family of AI modules: agents you configure with instructions, sources and tools; AI fields that fill themselves from a prompt; "Update with AI" in server actions; and AI hooks in Documents, CRM, Live Chat, Knowledge and Website. All of it is Enterprise-only, because every AI module depends on the base ai module, which is licensed OEEL-1. It calls OpenAI or Google Gemini, so a self-hosted server needs your own API key and PostgreSQL with pgvector.
This is part 5 of our Odoo 19 guide. It describes what the modules do according to their code in the 19.0 source, not the marketing pages.
The Odoo 19 AI modules and what each one does
| Module | What it does | Licence |
|---|---|---|
ai | Base: agents, topics, sources and embeddings, the AI server action type, Ask AI, AI drafting in emails and HTML fields, provider settings | Enterprise |
ai_app | The AI app and its menus: Agents, Topics, Settings, Default Prompts | Enterprise |
ai_fields | AI computed fields, filled from a prompt | LGPL, but needs ai |
ai_server_actions | "Update with AI" on Update Record server actions | LGPL, but needs ai_fields |
ai_documents | Automatic sorting of files into Documents folders | Enterprise |
ai_crm | A "Create Leads" topic so agents can create CRM leads | Enterprise |
ai_livechat | AI agents answering Live Chat visitors | Enterprise |
ai_knowledge | Knowledge articles as agent sources; AI drafting in articles | Enterprise |
ai_website | AI-written text for new website pages; AI Live Chat snippet | Enterprise |
Most of these install automatically when their parent apps are present. A helper module, ai_auto_install, installs the base ai module only if the pgvector extension is available on the database server, so if AI is missing on a self-hosted instance, check PostgreSQL first.
AI agents, topics and sources
The AI app (AI › Agents) is where you build agents. Each agent has:
- System Prompt: its instructions, tone and boundaries.
- LLM Model: the default is GPT-4o. The list covers OpenAI models from GPT-3.5 Turbo to GPT-5 and Gemini 1.5 and 2.5 models.
- Response Style: Analytical, Balanced (default) or Creative.
- Sources: files or URLs, plus Knowledge articles and Documents files when those modules are installed. Sources are indexed into vector embeddings, which is why pgvector is required. Tick Restrict to Sources to stop the agent answering from general knowledge.
- Topics: bundles of instructions plus tools. A tool is a server action marked Use in AI, so an agent can only do what you have explicitly given it an action for.
Odoo ships several system agents. Ask AI turns plain questions ("pivot of products with less than 5 in stock", "top 5 sales reps by revenue") into searches and opens the matching view. Odoo Agent is the general-purpose default used by the AI drafting prompts and by AI server actions. Under Configuration › Default Prompts you can edit the prompts behind buttons such as "Summarize the chatter conversation" or "Write a followup answer".
AI fields and "Update with AI" server actions
ai_fields adds an AI tab to field definitions: tick AI field and write an AI Prompt that can reference other fields on the record. With Studio installed, you can create these fields there instead of in developer mode. Typical uses are a one-line summary of a long description, a product category guessed from the product name, or a sentiment tag on a ticket.
Two behaviours are worth knowing. Empty text, char and HTML AI fields are filled by a scheduled action ("AI Fields: Compute AI fields") in batches, starting with the most recent 50 records when the field is created. That job currently only runs with an OpenAI key; if none is set, it skips silently. Other field types, such as dates, numbers and many2one or tag fields, are computed from the form view instead.
ai_server_actions adds Update with AI as a way to compute the value in an Update Record server action. Combine it with an automation rule and you get, for example, "when a lead is created, set its industry from the company description". Only stored fields can be updated this way. Separately, the base module adds an AI action type that runs a prompt with a chosen set of tools.
AI in Documents, CRM, Live Chat, Knowledge and Website
- Documents (
ai_documents). An administrator writes an AI Folder Sort Prompt on a folder. Files dropped there are sorted into the folders named in the prompt and tagged, using the tools "AI: Move To Folder" and "AI: Add Tags". The AI can only move files to folders the prompt mentions, and only ordinary files, not shortcuts. This is the most immediately useful AI feature for teams receiving scanned bills and HR paperwork. - CRM (
ai_crm). Adds a "Create Leads" topic. Its instructions tell an agent to offer a human follow-up when it cannot answer confidently, collect name, email and phone one at a time, confirm them, and create one lead. - Live Chat (
ai_livechat). Each Live Chat channel rule gains an AI Agent field next to the scripted chatbot. If both are set, the AI agent takes priority. With CRM installed, the default "Livechat AI Agent" also gets the lead creation topic. - Knowledge (
ai_knowledge). Knowledge articles can be added as agent sources, and AI drafting works inside articles. - Website (
ai_website). A system agent, "Website Page Generator", writes placeholder text for new pages in the language and tone you choose. It is archived by default, which keeps it out of the agent list, but Odoo calls it directly when generating a page. An AI Livechat snippet is also added to the website builder.
The useful question is not "where can we add AI?" It is "which repetitive judgement call does someone make fifty times a day?"
Sensible first uses and real limits
- Connect a provider. AI › Configuration › Settings: "Use your own ChatGPT account" or "Use your own Google Gemini account". Agree who pays the bill and set a spending limit with the provider.
- Start with Ask AI and email drafting. Low risk, visible benefit, no configuration.
- Add one AI field or one Documents sort prompt. Something a person would otherwise do by hand, where a wrong answer is cheap to fix.
- Only then build a customer-facing agent. Live Chat with Restrict to Sources on, a small, curated set of sources, and lead creation as the fallback.
The limits are real. Your data is sent to OpenAI or Google, so check this against your privacy policy and customer contracts; for Indian businesses, consider the Digital Personal Data Protection Act before sending personal data. Answers can be wrong, so keep a human in the loop anywhere money, stock or legal documents are involved. Costs scale with usage and are billed by the provider. And AI fields fill in batches on a schedule, not instantly.
If the built-in agents do not cover what you need, such as an AI step inside a custom workflow or a model running on your own infrastructure, that is custom work on top of Odoo. Our AI development team builds exactly that. Next in the series, part 6 sets up CRM.
Questions we get asked
Is AI available in Odoo 19 Community?
No. The base ai module in Odoo 19 is licensed OEEL-1, which is Enterprise only. The ai_fields and ai_server_actions modules are LGPL, but they depend on the base ai module, so they cannot be installed on Community alone. To use AI agents, AI fields or AI sorting in Odoo 19, you need an Enterprise subscription.
Which AI models does Odoo 19 support?
Odoo 19 supports OpenAI and Google Gemini. The agent model list includes GPT-3.5 Turbo, GPT-4, GPT-4o, GPT-4.1, GPT-4.1 Mini, GPT-5 and GPT-5 Mini, plus Gemini 1.5 Pro, Gemini 1.5 Flash, Gemini 2.5 Pro and Gemini 2.5 Flash. New agents default to GPT-4o. On a self-hosted server you enter your own API key in the AI settings.
Why does Odoo 19 AI need pgvector?
Odoo 19 turns agent sources such as PDFs, web pages and Knowledge articles into vector embeddings stored in PostgreSQL, and searches them by similarity when answering. That needs the pgvector extension, which Odoo documents as available for PostgreSQL 15 and up. Without it, the ai_auto_install module does not install the AI features automatically.
What is an AI field in Odoo 19?
An AI field is an ordinary field with a prompt attached. Odoo sends the prompt, including values from other fields on the record, to the AI model and stores the answer in the field. Text fields are filled in batches by a scheduled action that needs an OpenAI key, while other field types are computed from the form. You can create AI fields in Studio.
Can an Odoo 19 AI agent answer customers on Live Chat?
Yes. With the ai_livechat module, each Live Chat channel rule has an AI Agent field. Set it and visitors matching that rule talk to the agent first; if a scripted chatbot is also set, the AI agent takes priority. With CRM installed, the agent can collect contact details and create a lead when it cannot answer confidently.