AI is the headline of Odoo 20, but it's also the first Odoo release where AI isn't free. Every Odoo 20 AI feature agents, image generation, the website assistant now runs on purchased credits, according to the official Odoo 20 release notes. That changes the question from "what can this do" to "what will I actually use, and is it worth the cost."
This post explains what agents and skills do, what MCP connectivity actually means, and what you need to know about IAP credits before switching any of it on without the vague "AI will transform your business" framing that tends to accompany releases like this. If you want the broader picture of everything else that changed in this release, our Odoo 20 new features guide covers it; this post stays focused on AI.
What's Actually New: Agents, Skills, and MCP at a Glance
Feature | What it is |
AI Agents | Can create records, update records, answer questions about files, and be triggered automatically by scheduled or automated actions |
Skills | The renamed version of what Odoo previously called "topics" the instructions and tools that define how an agent behaves in context |
MCP | A protocol that lets Odoo connect to a database via a standardized method, described in the notes as "connect Odoo to anything" |
IAP Credits | All AI features now require purchased credits no AI feature runs for free by default |
AI Website Assistant | Can build pages, edit menus, and create or modify forms from a description |
Voice Interaction | Lets a user dictate requests directly to an agent instead of typing |
(Full detail on each of these, straight from Odoo, is in the AI section of the release notes.)
How Odoo 20 AI Agents Actually Work
An agent in Odoo 20 isn't a chatbot bolted onto the interface. According to the release notes, agents can directly create and update records on your behalf, and they can answer questions about a file while you're previewing it, rather than requiring you to describe the file's contents yourself.
One practical interface change: agents now show specific, live updates on what they're doing, replacing the older generic "AI is thinking" message. If an agent is pulling data, checking a record, or drafting something, you see that step rather than a blank loading state.
Agents aren't limited to manual chat requests either. They can be summoned by automated and scheduled actions, meaning an agent can run as part of a workflow without someone opening a chat window to trigger it. The release notes also describe agents as able to update themselves to best serve their use cases, a form of self-adjustment tied to how they're being used, though the notes don't go into the mechanics of how that self-updating works.
There's a built-in safeguard worth knowing about: if an agent reaches its maximum tool call limit, it prompts the user for confirmation before proceeding, rather than continuing to act indefinitely. Agents can also request permission or offer suggestions through interactive responses meaning you can respond to an agent's question or proposal by clicking a button rather than typing a reply.
A few other capabilities confirmed in the release notes: agents can filter by time period when building views (weeks, quarters, etc.), can be asked to reprocess sources to use the most up-to-date version of a file, and can send and receive files directly inside a chat conversation.
From Topics to Skills: What Actually Changed
It's worth being precise here: this is a rename, not a new capability. The release notes describe it plainly "Topics," the collections of instructions and tools that define how an AI agent behaves within a specific context, are now called "Skills." Functionally, nothing about how that configuration works is described as different; only the term changed.
If you've configured agent behavior in a previous version using topics, the underlying concept carries over; you're not being asked to rebuild anything, based on what the release notes confirm. The rename does suggest a framing shift, though: "skills" reads more like a deliberate capability an agent has, rather than a loose context a conversation happens to be about. If you're planning a more involved agent setup for your business, that kind of configuration work is something our Odoo customization team handles directly.
MCP: How Odoo Connects to External Tools
MCP (Model Context Protocol) is an open standard for connecting software to external tools and data sources in a consistent way, rather than building a custom one-off integration for every service. You can read the full technical specification in the official MCP documentation if you want the underlying protocol details.
What the Odoo 20 release notes confirm is narrower and more specific: Odoo can now connect to your database via MCP. Beyond that single confirmed capability, the notes don't list specific third-party integrations or prebuilt MCP workflows. In practice, this is best understood as Odoo opening a standardized door for IT teams to connect Odoo to other tools in their stack, rather than Odoo shipping a library of ready-made MCP integrations out of the box. If you're evaluating what that could look like for your specific systems, that's a scoping conversation worth having with an Odoo implementation partner before assuming a particular integration is supported.
The AI Website Assistant
The AI Website Assistant goes beyond a writing helper. Per the release notes, it can create new webpages from a description, edit a site's menu creating, renaming, moving, nesting, reordering, or deleting items and create or edit webforms.
It can also draw on outside input while building: it can take inspiration from other websites, pull from your own database records (such as products or events), and add interactive elements like counters, toggles, tabs, or animations. For images, it can source them from Unsplash or generate them directly.
One detail worth noting for editorial control: any building block that the AI Website Assistant has generated or modified is marked with an indicator in the Website editor, so it's visible which parts of a page were AI-produced versus manually built.
IAP Credits: What You Need to Know Before Turning AI On
This is the section to read before enabling anything. The release notes state it directly: all AI features now need credits, which can be bought via IAP. That's a structural change from any prior assumption that AI features in Odoo came at no separate cost; there's now a direct, ongoing expense tied to usage.
What the release notes do not specify is pricing: there's no stated cost per credit, per action, or per agent conversation. Before budgeting for a rollout, verify current IAP pricing directly through Odoo's official site rather than estimating from the release notes, since that information simply isn't part of what Odoo published in this release.
Our recommendation, based on general rollout practice rather than anything stated by Odoo: pilot with a single team or a single repetitive workflow first record creation from a specific form, for example before enabling agents broadly. That gives you a real usage pattern to estimate from, rather than committing to a credit budget based on guesswork. This is also the kind of planning work our team folds into a broader Odoo migration checklist conversation when AI adoption is part of a larger upgrade.
Other Notable AI Updates in Odoo 20
A few additional changes round out the AI section of the release notes:
- Automatic AI model selection — once you select an AI provider, Odoo automatically uses what it determines to be the best model for the task at hand.
- Stored conversations — conversations with agents are now stored for up to 30 days after the last message was sent. This is a data-retention detail worth knowing for admin purposes; the release notes don't address compliance frameworks like GDPR, so don't assume this detail answers any compliance question on its own.
- Image generation — agents can generate images and enriched buttons for websites and mailings.
- Default prompts with dynamic values — you can add context and dynamic values directly into default prompts.
- AI chat request options — a "+" menu lets you add extra options to a request before sending it in an AI chat.
- Preview cards in live chat — when an agent mentions a supported record in a live chat conversation, the chat can show a rich, clickable card instead of a plain record name.
Should You Turn On Odoo 20 AI Agents Right Away?
Reasonable candidates for early adoption:
- You have a specific, repetitive record-creation or record-update workflow where automation has an obvious, measurable use.
- Your team already has some comfort with automation or scheduled actions, so agent-triggered workflows aren't a completely new concept.
- You're able to monitor and budget for IAP credit usage as you go, rather than needing cost certainty upfront.
Reasonable candidates to wait:
- You don't yet have a specific use case "we should probably use AI somewhere" isn't a plan, it's a placeholder.
- You don't have visibility into what credit usage would look like for your volume of activity.
- You'd rather see how early adopters' usage and costs play out before committing your own budget.
Neither path is inherently right. The release notes don't suggest urgency here, and credits make the cost of experimenting real, not hypothetical which is exactly why a scoped pilot tends to beat a broad rollout on day one.
Want Help Scoping AI Adoption?
If you're trying to figure out which workflows are worth automating with Odoo 20's AI agents and what that would realistically cost Creyox can help you scope a pilot before you commit to a rollout. Get in touch if you want a second opinion before turning anything on.
Final Thoughts
Odoo 20's AI features are genuinely more capable than a chat window; agents that create records, respond to files, and run on schedules are real functional additions, not window dressing. But the credit requirement means adoption is now a budgeting decision as much as a technical one.
The most useful thing you can do before enabling anything is pick one concrete workflow, test it, and look at what it actually costs to run before deciding how far to take it. That's a more reliable path than enabling agents broadly and finding out the cost later. For the full picture of what else changed alongside AI in this release, our Odoo 19 vs Odoo 20 comparison covers the structural changes this post doesn't touch.
Frequently Asked Questions
AI agents in Odoo 20 can create and update records, answer questions about files, and be triggered by automated or scheduled actions. They show live updates on their actions and can request permission through interactive buttons.
Yes. According to the release notes, all AI features now require credits purchased via IAP. Exact pricing isn't published in the release notes, so check Odoo's current IAP rates before budgeting.
MCP (Model Context Protocol) is a standard that lets Odoo connect to your database via external tools. The release notes confirm this capability but don't list specific third-party integrations.
There's no functional difference. The release notes describe "skills" as the new name for what was previously called "topics" the instructions and tools that define an agent's behavior. It's a rename, not a new capability.
The release notes don't mention a free trial or credit allowance. All AI features require purchased IAP credits, so confirm current terms directly with Odoo before assuming any no-cost testing period exists.