Odoo 19 introduced the AI app: agents, sources, AI fields and automated actions. In Odoo 20, agents create records, are triggered by scheduled actions and connect to external AI assistants via MCP.
Based on the official Odoo documentation. Capabilities depend on the installed apps.
Ask a question in plain language via Ctrl + K: the AI opens a view, shows a report or improves text. Standard Ask AI does not change records.
An agent consists of a system prompt, a response style, skills (Topics) and sources. An agent without skills only provides information and takes no actions.
PDFs, web links, files in the Documents app and Knowledge articles are indexed. “Restrict to Sources” limits answers to these sources only.
A prompt is written into the field via Studio or properties: a product description, a comment summary, structured data from unstructured text. Empty fields are filled daily by a scheduled action.
The AI reads the record and picks one of the tools marked “Use in AI.” Business rules must live in the tool’s code.
AI Auto-sort by folder: moving, tagging, activities, bill creation.
Answers questions in the website chat. For a sales inquiry, a demo request or a low-confidence answer, it asks for contact details and creates a lead.
In record description and note fields, speech is converted to text and summarized. The audio is deleted after processing.
Chatter summaries, translation, reply drafts, AI in email templates.
Based on the Odoo 20 release notes and 20.0 documentation.
In Odoo 20, a scheduled action triggers the agent, and the agent works only with the skills it is allowed to use.
Odoo passes the agent a list of late deliveries and unanswered quotations.
For each record, it identifies the cause and the owner and reads the latest correspondence with the customer.
An activity is created for the responsible employee, a summary is posted to the Chatter, and a draft email to the customer is prepared.
Sending the email and changing deadlines require the employee’s approval.
An external assistant connected via MCP works with the access rights of the API key owner. If it connects with an administrator key, it gets write access to the entire database. That is why we create a separate technical user and grant it the minimum set of groups. Tools that create and update records are tested on a staging database first.
In most cases, Odoo’s built-in tools are enough. A custom agent is built when several systems or an Uzbek-language channel are involved.
| Goal | Built-in Odoo AI | Custom agent (API or MCP) |
|---|---|---|
| Website chat Q&A and leads | Live chat agent | Not needed |
| Sales via Telegram or Instagram | No built-in tool | Yes, via the Odoo API |
| Document sorting and draft bills | AI Auto-sort, Odoo 20 agents | For complex rules |
| Analytics and reports for management | Ask AI, Accounting AI assistant | When data from several systems is needed |
| Odoo + amoCRM + MoySklad together | No | Yes |
Measured separately for each agent.
| Metric | Calculation |
|---|---|
| Auto-generated drafts | Number of records created by the agent and the share approved |
| Incorrect actions | Agent records that were canceled or corrected |
| Answer accuracy | Share of correct answers on a test question set |
| Adoption | Number of employees using the agent per week |
Odoo works with the language model you select, and it understands Uzbek text. But quality has to be tested for each task, especially for voice input.
Yes, if the agent needs to create records, run from scheduled actions, or if you need MCP. We first review your modules and customizations, then draw up an upgrade plan.
Yes. We first review the database structure and access rights, then configure the AI. For our broader Odoo work, see Odoo implementation.
We review your processes and recommend a specific task, KPI and plan for your first agent.