Odoo 20 Is Officially Released: AI Agents, MCP and What to Know Before Upgrading

Odoo 20 is officially here. At the opening keynote of Odoo Experience 2026 in Brussels on 24 September 2026, Odoo unveiled the new version together with full release notes covering changes across more than 50 apps. In our earlier roadmap preview, Sonix summarized what Odoo planned to build. Now that the final release is out, this article checks the claims against reality: which features actually shipped, what changes you need to plan for when upgrading, and what companies should weigh before deciding.

A note on sources: Odoo publishes its release notes on its official website. This article draws on those notes as read and analyzed by several Odoo partners, plus Odoo’s launch video and the 20.0 branch on GitHub. Sonix has not yet run Odoo 20 on a live client project, so anything that reads as an assessment is labeled as our opinion, not an implementation result.

Overview

What stands out in Odoo 20

The most visible change is that the AI agent no longer only suggests: it can now create and update records. Alongside that come an MCP connection to outside AI tools, a new look with cleaner icons and softer colors, and a long list of improvements across Sales, Inventory, Manufacturing, Accounting, Website and POS.

It boils down to three groups:

• AI with write access: agents create and update records, read instruction files (such as a PDF) and run inside scheduled automations.
• MCP: connect your Odoo database to an outside AI assistant, which reads and writes data within the user’s access rights.
• Housekeeping in the core: simplified access rights, a new way of tracking changes in the chatter, parent accounts in the chart of accounts. None of this shows in a demo, but it is what you feel most when upgrading from version 17, 18 or 19.

AI

AI agents: from assistant to operator

In Odoo 19, AI mostly answered and summarized, and moving a change into the system was still a human’s job. In Odoo 20, agents can create and update records, and they can be called from automated and scheduled actions.

Here is the example from Odoo’s launch video. An order arrives by email. Instead of retyping it, the user gives the agent one instruction: whenever an order email comes in, create the sales order, send the customer a confirmation and flag it for my review. The agent explains what it is about to do, and from then on the process runs by itself.

Other improvements worth knowing in the AI area:

• Ask questions about a file while previewing it, and send files into an agent chat.
• A chat history for agents inside automations, showing how the agent handled a task and where it went wrong.
• A tool call limit: once it is reached, the agent asks the user before carrying on.
• “Topics” are renamed “skills”: the sets of instructions and tools that define how an agent behaves.
• Voice dictation of requests, and automatic model selection once you choose an AI provider.
• Agent conversations are stored for 30 days after the last message.

Sonix’s view: an agent with write access shifts the weight of a project onto permissions and data quality. If the agent’s limits are set deliberately and the data is clean, it works well. If you skip both, it just makes a mess faster.

Technical & Integration

MCP: connecting outside AI tools to Odoo

Odoo 20 lets you connect your database to an outside AI assistant through the Model Context Protocol (MCP). According to Odoo, the outside tool reads and writes ERP data within the access rights of the connecting user. In other words, the AI tool only sees and changes what that user is allowed to.

In the launch video, the database is connected once and the outside tool then works directly on Odoo data. This is the practical version of what the roadmap said about MCP.

For technical teams, the first job before switching it on is to review permissions: the account used for the connection should have limited rights, not administrator rights.

Sales & CRM

Sales and CRM: faster quotes, cleaner pipelines

• Quotation section templates: save a group of lines and reuse it, pick a quotation template right when creating an order, save any quotation as a template. Quantities per section are supported too.
• Order lines without a product: created from a description, with a setting that enforces a product where one is needed.
• Editable margins on the line, with the sales price recalculated. There is also line numbering, product images on quotes and a switch between prices with and without tax.
• Periodic pricelists (for example for certain weekdays), price rules per packaging type, and single-use discount codes.
• Commissions based on the amount actually paid, including manager commissions.
• Lead generation from the Dun & Bradstreet database (Odoo speaks of more than 580 million companies), filtered by segment and location. According to the release notes, this source replaces Clearbit.
• Pipeline switcher between sales teams, with new leads defaulting to the pipeline you are viewing.

Supply Chain & Manufacturing

Purchasing, Inventory and Manufacturing: the most changes for trading and manufacturing companies

Inventory and Purchasing

• Suggested minimum and maximum stock levels on reordering rules, calculated from demand history, desired days of coverage and order frequency.
• Vendor quality rating built from past orders, and an AI agent that points out vendors worth renegotiating with (both shown in the official video).
• An allocation report showing which customer orders are waiting for incoming stock, with reservation in one click.
• Inventory at a past date, a stock aging report, landed costs per product, and retroactive COGS updates when a cost changes after delivery.
• A CMR document for road freight, prefilled from the delivery.
• Better intercompany flows: one company resupplied from another’s warehouse, with valuation transferred correctly.

Manufacturing (MRP)

• Continuous Production: register produced quantities directly on work orders, and the next operation is released as soon as part of the quantity is ready, without closing the whole manufacturing order.
• A manufacturing order Kanban grouped by week, showing component status, active work center, deadline and remaining time.
• New Kanban and Gantt views for work orders, with drag-and-drop planning and a per-employee view.
• Bill of materials comparison, extra costs entered on the BoM, splitting an ongoing order, and resetting confirmed or cancelled orders to draft.
• MO cost: provisional while in progress, actual once closed. There is also one “Produce” button instead of the previous two.

Accounting & HR

Accounting and Payroll

Accounting

• Accounting AI agent: ask about receivables, gross profit or cash flow and get figures from your reports. In the video, the agent also audits a suspected bank reconciliation error and returns recommendations.
• Pay bills from Odoo: select a bill, click pay and confirm with your bank, individually or in bulk. Availability depends on your bank and country, so check what works in your market before promising it to a client.
• Bill line prediction from history, duplicate detection with two warning levels, and protection against double payments.
• Customer payment reminders, manual or automatic, with a simpler workflow.
• Intercompany: when a subsidiary invoices and the parent company receives the payment, Odoo reconciles it automatically.
• Batch actions on bank transactions, and batch reset to draft.

Payroll

• Net-to-gross simulation showing total cost, plus test pay runs with printed payslips before validation.
• A dashboard focused on setup warnings, salary indexation by percentage or fixed amount, and working schedules defined as hours per day.
• Expenses reimbursed through payslips are added to the salary computation automatically.

Website, POS & Field Work

Website, eCommerce, POS and field work

• Website built by prompt: the AI assistant creates pages, forms, menus and interactive blocks, rewrites copy, translates, and proposes SEO titles and descriptions. In the shop, AI edits a product’s name, price and description and generates images.
• Ready for AI search: an llms.txt file from the website settings, structured data added by default, cleaner sitemaps and responsive images for mobile.
• eCommerce: customer returns from the portal, minimum order quantity, preferred delivery date, progress bars toward free shipping, a range filter for numeric attributes, stock-based publishing and automated related products. Tax-included or tax-excluded display per pricelist serves B2C and B2B customers on one website.
• Offline mode: view, create and edit records without a connection, and changes sync automatically once you are back online. For service crews and field work, this is a real change.
• Field Service inside Planning, with travel time calculated automatically and the crew’s live location on a map.
• POS: a new dashboard, a redesigned floor plan editor, snoozing sold-out products, notes from self-ordering on the kitchen screen, several currencies at one checkout, and service fees.
• Odoo Phone, Discuss and Marketing Automation: buy phone numbers directly in Odoo, build call menus with drag-and-drop, get recordings and transcripts, and route calls by hours, location or team. Marketing Automation can build a campaign from scratch, from a template or by asking AI, with triggers based on events, dates, webhooks or on-demand runs.
• Timesheets: a new widget and an AI assistant help prevent forgotten hours, based on an ActivityWatch integration that tracks device activity. Odoo states that this data stays private.

Industry Verticals

Industry packages

Odoo 20 adds 13 new industry packages, including machine and tool rental, catering, vineyards, public institutions and engineer-to-order manufacturing. About a dozen existing packages, among them hotel, property developer and property management, were reworked.

Compared with the “about 20 industry versions” mentioned before launch, the reality is 13 new packages plus some existing ones upgraded. Companies in these industries should watch closely, because an industry package can shorten implementation compared with customizing from scratch.

Upgrading

What to watch for when upgrading to Odoo 20

Several changes in Odoo 20 remove or rebuild mechanisms that custom modules and reports rely on. On an existing installation, this is what sets the scope of the upgrade.

Beyond the table, here are five more things to check before upgrading:

• Localization: the release notes list localization changes country by country. In the sources Sonix reviewed, we did not find a dedicated entry for Vietnam, so e-invoicing, tax reports and bank formats should be tested on a copy of your data.
• AI costs: decide which processes deserve an agent and estimate credit usage.
• Payroll and expenses: the removal of work entries and the new place of employee expenses may change your current routines.
• Bank rules: manual journal entries on bank accounts are no longer possible without a bank transaction.
• Custom modules and hosting: every custom or third-party module must be migrated and tested. For self-hosted installations, the 20.0 source code requires Python 3.12 to 3.14 and PostgreSQL 16 or newer.

One more point on the support policy: Odoo maintains the last three major versions, so with Odoo 20 out, version 17 leaves support. Installations on version 16 and older were already outside it.

Roadmap

Should you upgrade right away?

The answer depends on where you are today:

• A new implementation starting now: start on Odoo 20. The first weeks after a launch are a period of fixes, so if you plan to go live within the next month, leave a buffer for release updates.
• Currently on Odoo 19: no rush. Support continues, so plan an upgrade when a new feature brings a concrete benefit, such as AI agents in automations or continuous production.
• Currently on Odoo 17 or 18: plan the upgrade within the next year. Version 17 has just left support, and version 18 will leave it with the next release.
• Currently on Odoo 16 or older: the upgrade is a project, not an update. Combine it with a review of which custom modules are still needed, since Odoo 20 covers some of them as standard.

The safest approach has not changed: upgrade a copy of the database first, test your real processes with key users, and only then move production. A new major version is also a good moment to clean up processes, not only to switch version numbers.