AI features

Odoo Dashboards has four AI features. Each one starts as a conversation in plain language, and each one builds or answers with the same datasets, widget types and filters you would use by hand.

Feature	Where you find it	What it does
Design a dashboard	Dashboard Builder → New Dashboard → Design it with AI (optional)	Plans a whole dashboard with you, then builds every widget on your existing datasets.
AI visualization	Dashboard Builder → item → Visualization Type → AI	Designs one widget with you, then creates it as a normal tile, chart, table or matrix.
Refine with AI	Dashboard Builder → a saved item → General tab	Talks through changes to an existing widget and applies them when you say so.
Analyst	Any dashboard → round button in the bottom-right corner, Ask AI about this data	Answers questions about the dashboard’s data and shows the queries it ran.

The result of every builder feature is an ordinary widget. You can open it, change any setting by hand, and save it like a widget you built yourself.

Who can use them

The AI features need the AI Analytics (SAAS) access right on the user (Settings → Users, developer mode, see Install the app). Settings administrators have it automatically. People without it don’t see any AI button, card or box.

Give this right only to people you trust with your data. AI-designed widgets and the Analyst read the dataset views directly, and dataset views don’t apply Odoo’s record rules or company restrictions. Someone who can only see their own sales orders in Odoo can see everyone’s through a dataset.

The builder features also need the usual Dashboard Builder (SAAS) right, because they work inside the Dashboard Builder.

An administrator can go further with AI governance: give each person a credit allowance, limit the datasets they and the AI may use, and limit what they may do to dashboards.

Nothing changes until you press the button

Every AI feature that changes your work follows the same rule: the AI proposes, you confirm.

Chatting never creates or changes anything. You can ask questions, push back and change your mind as often as you like.
The AI shows its plan as a card: the dashboard and its widgets, the one widget, or the exact list of changes to a widget.
Only a button builds or applies that plan: Build this dashboard, Generate this visualization or Apply changes. Discard drops a proposed change.

The Analyst only reads data, so it has no button to confirm. It never changes a dashboard.

In every AI chat box, Enter sends and Shift+Enter starts a new line. If a message fails to send, it goes back into the box so you can try again, and an error message appears in the corner.

Credits

AI requests use credits. Every request is priced by how much work it took, so a chat message costs less than building a whole dashboard. Each chat turn, each build, each refinement and each Analyst question uses credits, and a request that fails after the AI has already answered still uses them.

Your company buys credits as one pool. On its own, everyone with the AI right draws from that pool. The AI visualization card shows what is left (Credits remaining) with a Buy credits link. With AI governance, an administrator can give each person their own allowance out of the pool. See AI governance and credits.

When the pool is empty, every AI request is refused with You have no AI credits left. Please top up to keep using AI.

What is sent to the AI service

AI requests go from your Odoo server to the Odoo Dashboards service, which passes them to Anthropic’s Claude models. Your Odoo database is never opened to the service; your Odoo server sends what is listed below and receives an answer.

The builder features (Design a dashboard, AI visualization, Refine with AI) send:

what you typed in the chat, and the earlier messages of the same conversation;
the names of the datasets you may use, the Odoo model each one is built on, and for each column its name, its type and the Odoo field it comes from;
the names of your filter operations (such as Is between or Last 30 days) and which column types they apply to;
which columns of different datasets can be joined to each other;
for Refine with AI, the widget’s current settings, including its name, its columns and its filters with their saved values (for example a status filter set to Sales Order), and the list of refinements already applied to it.

They don’t send any records or rows from your database. The dataset’s SQL is not sent either.

The Analyst sends:

your question and the earlier messages of the conversation;
the dashboard’s name and the names and types of its widgets;
the datasets behind the dashboard, with the same column details as above, and how they join;
the results of the queries it runs. The Analyst writes read-only queries, your Odoo server runs them on the dashboard’s datasets, and up to 50 result rows per query go back to the AI so it can answer. Those rows contain real values, for example customer names and amounts when you ask about customers.

Your own database keeps a copy of every completed AI request: administrators can read the prompts, the AI’s answers and the cost in the AI Generation Log.

Where there is no AI

The AI doesn’t create or change datasets. It builds only on datasets that already exist, so if the data you ask about isn’t in any dataset, it tells you what is missing and someone with the Dataset Builder (SAAS) right adds it. There is no AI in the Dataset Builder, the join editor or the custom field wizard.

The AI can combine two or more existing datasets for one widget, the same way joins do. When it proposes this, the plan says Combines: followed by the datasets.

The AI doesn’t set up global filters, the layout, sticky notes or targets.