Odoo 20 AI Agents: How to Build Them and Where to Use Them

An Odoo 20 AI agent is an assistant you build inside Odoo's AI app. You give it a system prompt that describes its job, skills that let it look up, create or update records, and sources such as PDFs, web pages or Knowledge articles to answer from. Your team can then talk to it from any screen, your website visitors can chat with it, and an automation rule or scheduled action can start it with nobody asking. For everyday tasks you can build an agent without writing code. You need Odoo Enterprise on a plan that includes the AI app, and every request uses Odoo AI credits, which you buy separately.

Below you'll find how an agent is put together, how to build one, where it helps in each department, and the costs and risks to plan for. For the rest of what changed in this release, see our article on Odoo 20 new features and upgrading.

What an AI agent is in Odoo 20

Odoo describes an agent as a virtual assistant that understands plain language and can act through the tools you give it. Each agent is a record in the AI app with a few parts.

• System prompt: The agent's job description. It says who the agent is, what it handles, what it must never do and how it should answer. It applies to every conversation the agent has.
• Response style: The tone of the answers. Odoo 20 offers Standard, Nuanced, Snappy & Creative, and Slow & Rigorous. For business data, Standard or Slow & Rigorous are the safer choices.
• Skills: Each skill bundles written instructions with one or more tools, which are the actions the agent can take. Without skills, an agent can only give information. It can't create or change anything.
• Sources: The material the agent reads before it answers, such as PDF files, web links, files from the Documents app and Knowledge articles. Odoo indexes them so the agent can find the right passage.
• Restrict to Sources: A checkbox that makes the agent answer only from its active sources. Use it for agents that talk to customers.

Odoo 20 comes with these preconfigured skills: View Builder (opens the right list, chart or pivot for a question), Information retrieval, Update Records, Create Records, Generate Images, Web Search, Website Builder, and Create Leads when the CRM app is installed. The tools available depend on your installed apps.

Where an agent can work

Your team opens an agent with the AI icon at the top of any screen or with Ctrl+K, and can attach files, dictate by voice or move the conversation into Discuss. An agent can also answer website visitors in live chat, open on a specific kind of record through a default prompt, or run from an automation rule or scheduled action. AI tools outside Odoo, such as Claude, ChatGPT or Codex, can connect through Odoo's MCP server and use the tools you expose to them.

What changed since Odoo 19

If you tried agents in Odoo 19, five changes matter. Topics are now called skills. Agents can create and update records, including from a file of instructions such as a PDF. Automated and scheduled actions can start an agent. In the chat, you answer an agent's questions and permission requests by clicking buttons, and it asks before going past its limit of tool calls. Odoo 19 also let you pick the AI model and plug in your own OpenAI or Google key. Odoo 20 runs every request through Odoo's own service, paid in credits, and Odoo's privacy policy states that for version 20 it chooses the AI provider itself.

How to build an AI agent in Odoo 20, step by step

Pick one narrow task before you open the app, such as "answer delivery questions on the website". An agent with a vague job gives vague answers and is harder to test.

1. Install the AI app. If your subscription is on the Standard plan, installing it moves you to the Custom plan, so check the price with your account manager first. Make sure the database has AI credits.
2. Create the agent. In the AI app, click New, enter an Agent Name and a short description.
3. Choose a response style that suits the audience.
4. Write the system prompt. Describe the role, the tasks, the limits and the format of the answers. Odoo's own preconfigured agents are good examples to copy from.
5. Add skills. Start with Information retrieval and View Builder for an agent that reads and reports. Only add Create Records or Update Records once you trust its answers.
6. Add sources. On the Sources tab, click Add a source and upload a file, paste links (one per line), or pick files from Documents or articles from Knowledge. Wait until each status changes from Processing to Indexed, then switch it to Active.
7. Decide on Restrict to Sources. Tick it when the agent must stick to your documents.
8. Test it. Click Test and ask the questions your team and customers really ask, including requests it should refuse.
9. Put it to work in one place first: a live chat channel, a default prompt on a model, or an automation rule.

Here's an example system prompt for an internal sales assistant. Adapt the names and rules to your business.

You are the sales assistant for our sales team.
You help salespeople find information about customers, quotations
and the pipeline, and you prepare follow-up emails.
- Use only data you can read in Odoo. If you can't find something, say so.
- Never change prices, discounts or payment terms.
- Never send an email yourself. Write a draft and show it to the user.
- Amounts are in the customer's currency, with the currency code.
- Answer in short paragraphs, in the language the user writes in.

Connecting the agent to the rest of Odoo

• Website live chat: In the Live Chat app, open a channel's configuration, go to the Rules tab, open or add a rule and choose the agent in the AI Agent field. If a channel has both a chatbot and an AI agent, the agent takes priority. Odoo adds a ready-made Livechat AI Agent that answers questions and, for requests such as custom pricing or contracts, collects contact details and creates a lead.
• Default prompts: In AI, Configuration, Default Prompts, create a rule, choose the models (for example Helpdesk Tickets), the agent to call and the instructions for that context. Add buttons such as "Summarize this ticket" so users get the same result every time.
• Automation rules: In Studio, open Automations, create a rule, choose the trigger and conditions, then select an AI agent and write instructions for what it should do. Type / to insert fields from the record.

An example of instructions for an automation rule that runs when a new ticket is created:

Read the customer's message on this ticket.
Set the ticket type to Question, Bug or Billing.
Set the priority to High only if the customer says their work has stopped.
Write a three-line summary in the internal note.
Don't reply to the customer.

Where to use AI agents, by department

For each area, we separate what Odoo 20 ships, what you can set up yourself and what needs a developer.

Sales and CRM

• Ships with Odoo: The Livechat AI Agent with the Create Leads skill, AI prompts in email templates that write a personal follow-up for each record, and questions about your pipeline answered as lists or charts.
• You can set up: A sales assistant like the one above, and an automation rule on new leads that summarizes the request and fills a field such as the industry or the product of interest.
• Needs development: Scoring leads with data from outside Odoo, or rules about discounts and approvals that the agent must never be able to break.

Website, live chat and helpdesk

• Ships with Odoo: AI answers in live chat, clickable preview cards when the agent mentions a supported record, similar ticket detection with suggested answers, and AI fields that summarize long customer messages.
• You can set up: A support agent that answers only from your Knowledge articles and product guides, with clear rules for when to hand over to a person, and a default prompt on tickets with buttons for a summary or a draft reply.
• Needs development: Answers that depend on data in another system, such as a carrier's tracking or an outside online shop.

Accounting and finance

• Ships with Odoo: Vendor bill digitization, which reads scanned or PDF bills and fills in the bill for review (a separate paid service), AI sorting and renaming of files in Documents, where a folder rule can create vendor bills, and answers about your finances based only on the accounting data in your database.
• You can set up: A finance assistant that answers "which customers owe us more than 60 days?" with the right report, and a Documents inbox prompt that sends bills, contracts and receipts to the right folder.
• Needs development: Any tool that posts entries or validates payments with your own checks built in. We'd keep a person on every posting.

Inventory and purchasing

• Ships with Odoo: The View Builder skill, which answers questions such as "show the last purchase order from this supplier" or "which products are late for delivery?" opened as filtered lists.
• You can set up: A scheduled agent that looks at products below their minimum each morning and leaves an activity for the buyer. Keep the buyer as the one who confirms orders.
• Needs development: Reading supplier price lists in their own formats, or connecting to supplier portals and EDI.

HR and recruitment

• Ships with Odoo: Resume digitization through the Documents Digitization service, AI email templates for interview invitations, and voice transcription with summaries for interview notes.
• You can set up: An HR assistant that answers staff questions on leave and expense policies from your own documents, with Restrict to Sources ticked, and an AI field on applicants that summarizes the CV against the job description.
• Needs development: Anything that ranks or rejects candidates on its own. Leave hiring decisions to people, and check local rules on automated decisions about candidates.

What businesses gain from Odoo 20 AI agents

The main gain is time spent on repetitive reading and typing. Tickets arrive sorted and summarized, vendor bills arrive filled in, and a manager who asks for sales by country this quarter gets the chart without building a report. The agent works inside Odoo with each user's own access rights, so there's no second system to keep in sync. Website visitors get answers at night and on weekends, and the requests that need a person reach your team as leads with the contact details filled in.

Limits, costs and good practice

Credits and plan

All AI features in Odoo 20 need IAP credits. At the time of writing, Odoo's IAP page sells Odoo AI credits at €1 each in packs of 10 to 1,000 credits, and says the amount used per request depends on the feature, the task and the size of the request. A scheduled agent uses credits on every run, whether anyone reads the result or not, so start with a low frequency and watch the balance for a few weeks. Odoo's pricing page lists agentic AI with Studio under the Custom plan.

Wrong answers

An AI model can state something false with confidence. Give the agent good sources, tick Restrict to Sources for customer-facing agents, and write in the prompt what it should do when it doesn't know, for example hand over to a person. When a source changes, reprocess it so the agent stops quoting the old version.

Access rights

Odoo's privacy policy states that when an agent reads Odoo data, it does so within the permissions of the current user and the agent's configuration. Review your access rights before you switch agents on, especially for portal users and for staff who shouldn't see salaries or margins.

Actions without a person

Odoo's documentation on AI server actions explains that the AI only chooses which tool to call, and the tool then runs unless its own code stops it. Business rules such as "never above this amount" belong in the tool, written by a developer. Give each agent the fewest skills it needs, let it prepare drafts for a person to send, and read its automation chats, where Odoo 20 shows how the agent handled each run.

Privacy

According to Odoo's IAP privacy policy, a request sends your prompt, the relevant conversation, the record you're working on and excerpts from the agent's sources to an AI provider, currently OpenAI or Google Gemini, chosen by Odoo. Odoo says the providers don't store this data or use it to train AI models. In the database, conversations with agents are kept for up to 30 days after the last message. If your contracts or local law limit where client data may go, check this with your legal adviser before you add sensitive documents as sources.

Test on a copy first

Build and test new agents on a duplicate or staging database, with a test user who has the same rights as the real users. Try requests it should refuse as well as the normal ones. Then go live with one agent and one process, and add the next one once the first has run for a few weeks.

When you need more than the standard

The standard skills cover most reading, reporting and drafting work. For the cases below, you'll need a developer or another tool.

• Custom tools: A tool is a server action marked Use in AI. A developer can write tools for your own processes, with your business rules in the code, and describe their inputs so the agent knows how to call them.
• Other systems: When an answer depends on another platform, a carrier or a bank, that connection has to be built, in code or with a workflow tool such as n8n.
• Your own AI model: Odoo 20 chooses the model for its built-in agents. If your company has its own AI subscription, an AI tool that supports MCP can connect to your database from outside, within the user's access rights and the tools you expose.
• WhatsApp: Our own AI agent, built on Orbit, answers our customers on WhatsApp 24/7 in any language. Orbit can do the same for your business and, depending on your setup, work with your Odoo data. For custom agents, n8n workflows and integrations, see our AI agents and automation page.

FAQ

Are AI agents available in Odoo Community?

No. The AI app and its agents are part of Odoo Enterprise, and Odoo lists agentic AI under its Custom plan. Community users can connect outside AI tools through custom modules or automation platforms such as n8n.

How much do Odoo 20 AI agents cost?

You pay for the plan that includes the AI app, plus Odoo AI credits for every request. Agents that run on a schedule or on many records use the most credits.

Can I use my own ChatGPT, Gemini or Claude account?

Not for Odoo's built-in agents in version 20, where Odoo selects the provider and charges credits. Odoo 19 allowed your own OpenAI or Google key. In Odoo 20 you can connect your own AI tool to the database through Odoo's MCP server, using OAuth or an API key.

Is my business data used to train AI models?

Odoo's IAP privacy policy says it isn't. The data still leaves your database while a request is processed, so choose carefully which documents you add as sources.

Can an AI agent change records without approval?

Yes, if you give it the Create Records or Update Records skill or a tool that writes data. In a chat, Odoo 20 can ask the user for permission through buttons. In automations nobody is watching, so keep those agents to the smallest set of skills and build hard limits into their tools.

Do I need a developer to build an AI agent in Odoo?

Not for agents that answer from your documents, open reports, summarize records or run on an automation rule with the standard skills. You need a developer for custom tools, connections to other systems and rules that the agent must never be able to break.

Plan your first Odoo AI agent with NavyBits

NavyBits has been an Odoo Silver Partner since 2016, with 6 Odoo-certified experts, and a senior engineer leads every project. We can set up Odoo 20 agents on your processes, write the custom tools and access rules they need, and connect them to your other systems. Our AI automation assessment starts from $1,000 and ends with a written report on which tasks suit an agent, what it would cost to run and what to build first. See our Odoo services and AI agents pages, or book a free 30-minute call to talk about the first agent for your business.