Where Odoo 19 left off
Odoo 19 (September 2025) was the first release with a dedicated AI app. According to the Odoo 19 release notes, it already offered:
- AI agents that answer from your documents and query database records in plain language.
- AI fields that fill themselves in, and AI steps in server actions.
- "Ask AI" search that turns a question into an Odoo filter.
- Voice transcription with summaries, AI drafting in emails and the chatter.
- An AI agent for website live chat that can create leads.
- A choice of OpenAI or Gemini models. On Odoo.sh and on-premise you had to bring your own API key.
In practice, Odoo 19 AI was mostly an assistant: it read, summarised and suggested. Odoo 20 lets it act.
What is new in Odoo 20 AI
The AI section of the Odoo 20 release notes lists these changes:
Agents that create and update records
- Create records on request, including from an uploaded file of instructions such as a PDF.
- Update records on request.
- Run from automations: agents can be called by automated and scheduled actions, not only from a chat window.
- Self-update: an agent can adjust its own configuration to serve its use case better.
- Subagents: an agent can hand work to more specialised agents (agents documentation).
- Tool call limit: when an agent reaches its maximum number of tool calls, it asks the user before going on.
Configuration and chat
- "Topics" are renamed Skills: sets of instructions and tools that define what an agent may do. Per the documentation, an agent without Skills can only provide information.
- Agents can answer from Sources (files, web links, Documents, Knowledge articles), with an option to restrict answers to those sources.
- Interactive replies: users answer an agent's questions and permission requests with buttons.
- Voice dictation to agents, image generation, and file uploads inside agent chats.
- Conversations with agents are stored for up to 30 days after the last message.
AI inside individual apps
Odoo 20 also adds AI to specific apps. From the release notes:
| App | What AI does in Odoo 20 |
|---|---|
| Accounting | Answers questions about company finances, based only on the accounting data in the database. |
| Documents | Renames uploaded files. |
| Discuss | Records calls and summarises what was said. |
| eCommerce | Edits product names, prices and descriptions, and generates product images, from the shop page. |
| Helpdesk | Finds similar tickets to suggest answers. |
| ESG | Assigns the most suitable emission factor to emissions. |
| Marketing Automation | Helps set up campaigns. |
| Social Marketing | Writes social media posts. |
| Website | An AI Website Assistant that edits blocks, creates pages, forms and menus, improves page SEO and can copy the layout of a page from a URL. |
Most of these apps (Accounting with full reports, Documents, Helpdesk, ESG, Marketing Automation, Social) are Enterprise-only, so their AI features are too.
The MCP server: connecting outside AI tools
MCP (Model Context Protocol) is an open standard that lets AI assistants call tools in other systems. Odoo 20 exposes one per database. According to the MCP server documentation:
- The endpoint is your database URL plus
/mcp. - Access is by OAuth (recommended, internal users) or by an API key with the "MCP" scope, created under My Preferences → Security. Deleting the key revokes access.
- Five read tools are exposed by default (get models, get fields, initial context, search, read group). Any other server action must be explicitly marked "Available in MCP"; a "Readonly Tool" flag lets a tool run without asking the user.
- The documentation covers clients such as Claude, ChatGPT, Codex and Grok, and warns that enabling dynamic client registration lets any application register as a client.
This is the most useful change for companies that already use an AI assistant or an automation platform: the assistant can read Odoo data and run approved actions without a custom API layer. Odoo's pricing page lists "Odoo Studio / Agentic AI" under the Custom plan, so check your plan before you design around it.
Pricing: AI now runs on credits
The release notes are explicit: "All AI features now need credits, which can be bought via IAP." The Odoo AI service page lists:
- EUR 1 per credit, in packs of 10 to 1,000 credits, with 20 free credits to start.
- Credits do not expire and belong to one database; auto-refill is optional.
- No per-feature price: usage "varies based on the specific AI feature, task complexity, and the size of your request".
Two practical consequences. First, the Odoo 20 documentation no longer describes bringing your own OpenAI or Gemini key, which Odoo 19 required on Odoo.sh and on-premise; AI requests go through Odoo's IAP service instead. Second, the AI documentation notes that installing the AI app on the Standard plan triggers an upsell. Budget credits by measuring real usage on a test database before you switch AI on for everyone.
Community edition: what you get without a subscription
We checked the official Odoo 20 Community source code. It contains no AI app, no agents, no AI fields and no MCP server (API keys in Community only have the RPC scope). What Community does include are small helpers that call Odoo's text service:
- AI translation in the HTML editor, the mail composer and the website builder.
- AI suggestions in the website setup wizard (titles, industry, theme).
- A hand-written
llms.txtfile for AI crawlers, edited in website settings.
The general "write or improve text with AI" editor command, which Community 18 had, is no longer in Community 19 or 20. So on Community, AI in business processes means a custom integration: a module or an automation that calls an AI model through its API and writes results back to Odoo. That is what our AI integration service covers.
Data and control: what to check first
- Retention: agent conversations are kept for up to 30 days after the last message.
- Processing: Odoo's public documentation and privacy policy do not state which AI providers process the requests or where. If that matters for your compliance, ask Odoo for it in writing.
- Permissions: an agent can only act through the Skills and tools you give it. Keep business rules in the tools' code, as the server-action documentation advises, not only in prompts.
- MCP exposure: only mark the server actions you really need as "Available in MCP", prefer read-only tools, and leave dynamic client registration off unless you need it.
Should a small or mid-sized company use it?
Our practical advice for an Odoo 20 Enterprise rollout:
- Start with agents that only read: answers from your documents, Helpdesk suggestions, the live chat agent with human escalation.
- Pick one repetitive workflow for a write-enabled agent (for example, creating draft records from incoming documents) and keep a human approval step.
- Use the free credits on a test database to estimate the monthly credit spend before rollout.
- If you already work with an external AI assistant, evaluate the MCP server before building custom API connectors.
- On Community, or for logic Odoo AI does not cover, plan a custom integration instead.
If you are deciding between upgrading to Odoo 20, staying on 19 or adding AI to Community, see our guides on migrating to Odoo 19 and Community vs Enterprise.