Odoo 20: What’s New? The Features, AI Changes & Upgrade Impact That Matter

Odoo 20 was released in September 2026. The important story is bigger than a list of new buttons: Odoo is becoming more agentic, more connected to external AI, more resilient when connectivity disappears, and more consolidated across operational workflows.

For an existing Odoo customer, the useful question is therefore not only “What is new in Odoo 20?” It is “Which changes affect our users, custom modules, integrations and upgrade plan?”

This guide focuses on those changes.

Octivex engineering note: Odoo 20 being released does not automatically mean every existing production database can be upgraded to it immediately. Upgrade availability and the exact process depend on the source version, hosting model, custom code and Odoo’s current upgrade tooling. Verify the exact database path before planning a production cutover.

Odoo 20 at a glance

The changes we would evaluate first are:

• AI Agents built around Skills, Sources and Tools.
• AI server actions that let AI select controlled business tools.
• An Odoo MCP server for supported external AI clients.
• Indexed company knowledge that agents can use as Sources.
• Expanded offline workflows.
• A Field Service flow integrated into Planning.
• Broad changes across Accounting, CRM, Sales, Inventory, Manufacturing, HR, Website, eCommerce, POS and Services.
• Framework-level changes that custom-module developers need to review.
• Continued migration toward the JSON-2 external API.

Rather than treating every release-note bullet equally, the sections below explain what these changes mean in practice.

1. AI in Odoo 20 is moving from answering to acting

The most important architectural shift in Odoo 20 is AI.

Odoo’s current AI documentation describes agents around Skills and Sources, while tools give the agent controlled capabilities. A Skill defines instructions and available tools for a task. Sources provide indexed information the agent can retrieve.

That makes a different class of workflow possible.

A basic assistant might answer:

“Summarize this opportunity.”

An agentic workflow can instead reason about the request, select an approved tool and continue a business process.

That difference matters because AI is moving closer to operational ERP data.

AI server actions: Manager and Worker

Odoo 20’s AI server-action documentation describes a useful separation of responsibilities.

The AI server action behaves like a Manager: it interprets context and decides which tool to use. The underlying tool behaves like a Worker: it performs the actual operation.

This is a good production pattern because business rules should remain deterministic.

If an agent is allowed to create, update or trigger something important, the underlying tool should still validate permissions, required fields and business constraints. A clever prompt should never become a replacement for application-level validation.

What an implementation team should define

Before enabling an AI-driven workflow, document:

1. what the agent is allowed to decide;
2. which tools it can invoke;
3. which user permissions apply;
4. which rules must remain deterministic;
5. which actions require human approval;
6. how the workflow will be evaluated after prompts, models or source documents change.

The valuable part of Odoo 20 AI is not simply that an LLM can generate text inside an ERP. It is that AI can participate in controlled workflows.

2. Odoo 20 MCP connects external AI clients to Odoo

Odoo 20 now documents an MCP server for connecting compatible external AI clients directly to an Odoo database.

The endpoint follows this pattern:

https://your-database.example.com/mcp

Authentication uses an Odoo API key. Current Odoo documentation includes configuration examples for clients including Claude Code and Codex.

The important security detail is that an MCP connection is not supposed to mean “give an AI unrestricted database access.”

Odoo exposes a limited default tool set, and additional server actions must be made available to MCP. A server action can also be designated a Readonly Tool when it does not write or modify records.

MCP and a conventional API are not the same thing

MCP is useful when an AI client needs to interpret a natural-language task and choose from approved tools.

A conventional API is better when software needs deterministic system-to-system behavior.

For example:

Good MCP-shaped task: “Find overdue opportunities with no recent activity, summarize the situation and use the approved follow-up tools.”

Good API-shaped task: “When shipment 1842 is completed in system A, update shipment 1842 in Odoo.”

Odoo 20 gives developers both directions, but they solve different problems.

3. Company documents can become AI Sources

Odoo’s current AI Agent documentation lists Sources including:

• PDF files;
• web links;
• Documents-app documents;
• Knowledge articles.

These sources can be indexed and used by an agent when it needs relevant information. Odoo also documents a Restrict to Sources option for constraining an agent to active sources.

That makes internal knowledge use cases much more practical.

A company can potentially ground an agent in HR policies, SOPs, product manuals, support documentation, sales playbooks and internal Knowledge articles.

This overlaps with what many teams call RAG, or retrieval-augmented generation.

It does not mean every custom RAG project should disappear. Requirements around retrieval control, external datasets, model choice, scale, evaluation or security may still justify a custom architecture. But in Odoo 20, native Agents and Sources should now be evaluated before building a separate knowledge assistant from scratch.

4. Offline mode may matter more than the AI headline for field teams

Odoo’s Odoo Experience 2026 material demonstrates an Offline Mode in version 20. The documented session covers prerequisites, a dedicated offline UI, browsing, searching, creating and editing records, archiving/unarchiving, deleting, and synchronization through a queue.

For teams working in warehouses, customer sites, construction environments, remote facilities or field service, this can be operationally significant.

However, “offline mode” should not be interpreted as “every Odoo workflow works identically without connectivity.”

A production test should verify:

• what records are available before disconnection;
• which searches work;
• what users can create and edit;
• how queued changes synchronize;
• how conflicts are handled;
• how custom modules behave;
• whether attachments or external services are required by the workflow.

For a field deployment, those answers matter more than the feature headline.

5. Field Service is now closer to Planning

Odoo’s own Odoo Experience 2026 material describes the new Field Service flow as built into the Planning app.

The showcased flow brings together technician planning with areas such as serial-number history, linked technical documents, WhatsApp communication, stock by truck and offline mobile tools.

This is a good example of why an upgrade project should not simply port every customization forward.

If your current database contains custom Field Service or Planning logic, ask:

Was this customization built because the standard product could not handle the requirement, or is the requirement genuinely unique to our business?

If Odoo 20 now covers the old gap natively, removing obsolete customization may be better than migrating it.

6. The rest of the business suite is changing too

Odoo Experience 2026 includes dedicated Odoo 20 sessions across Accounting, CRM, Website, POS, Projects, Helpdesk, Rental, Subscriptions, Inventory, Manufacturing, Field Service/Planning and other areas.

For an upgrade, the right approach is not to memorize every new feature. Build a regression plan around the workflows your company actually uses.

Finance

Test company and currency context, taxes, journals, reconciliation, analytic dimensions, approvals, multi-company behavior and auditability. AI-assisted finance features do not reduce the need for financial controls.

CRM and Sales

Inventory existing automations, Studio rules, scheduled actions, assignment logic and connectors. Determine whether Odoo 20 now covers any of those requirements natively before migrating custom logic.

Inventory and Manufacturing

Test end-to-end flows: replenishment, purchasing, receiving, picking, packing, shipping, lots and serials, routes, inter-company stock, BoMs, manufacturing orders, subassemblies, quality and valuation.

A custom manufacturing workflow can appear correct in the UI while producing a wrong downstream stock or accounting result. Validate the resulting data, not only the screen flow.

Website and eCommerce

AI-assisted creation can accelerate the starting point, but production quality still depends on information architecture, SEO, accessibility, performance, analytics, multilingual content and conversion design.

7. Developers have real Odoo 20 framework changes to review

Odoo 20 is not only a functional release.

The official ORM changelog already documents framework changes including:

• removal of Model._table_query;
• an updated RPC representation for binary fields using BinaryValue.filename;
• translation-related changes;
• a new populate script for generating test data;
• changes to storage of module-manifest translation terms.

These are the kinds of changes generic “Top Odoo 20 Features” articles often miss.

Changing a module’s manifest version and fixing the first traceback is not a migration strategy.

A custom-module review should cover at least:

• manifest and dependencies;
• Python models and ORM usage;
• computed and inverse fields;
• SQL/reporting models;
• XML views;
• actions and menus;
• ACLs and record rules;
• controllers and external endpoints;
• JavaScript/Owl code and assets;
• reports;
• scheduled and server actions;
• integration payloads;
• automated and manual tests.

Where applicable, test the module against both a clean target-version database and the upgraded customer database.

8. Be careful with outdated Odoo API information

There is already stale information online about Odoo’s legacy APIs.

Odoo’s current documentation says the XML-RPC and JSON-RPC APIs at /xmlrpc, /xmlrpc/2 and /jsonrpc are deprecated. The db service was removed in Odoo 20. The common and object services are currently scheduled for removal in Odoo 22, with the External JSON-2 API acting as the replacement.

That timeline matters because older material originally expected broader removal in Odoo 20.

If you maintain a Flutter application, portal, middleware service, BI integration or another external system connected to Odoo, verify the current official documentation instead of planning from an old conference slide or blog post.

Also remember:

• JSON-2 is an application-facing external API.
• MCP exposes controlled tools to compatible AI clients.

They both connect external systems to Odoo, but they are not interchangeable.

9. Odoo 20 vs Odoo 19: what deserves attention

The common theme is that Odoo 20 changes how work is initiated, how company knowledge is retrieved, how AI interacts with ERP operations, and how users continue working when connectivity or application boundaries would previously interrupt the process.

10. Should you upgrade to Odoo 20 now?

Do not decide from the feature list alone.

An Odoo 20 upgrade is worth evaluating when the new functionality solves real operational problems, your current version requires modernization, or maintaining old customizations and integrations is becoming more expensive than moving forward.

Before scheduling production, answer five questions:

1. Is Odoo’s upgrade path currently available for our exact source version and hosting model?
2. Which custom modules must be migrated, replaced or removed?
3. Which third-party apps are compatible with Odoo 20?
4. Which external integrations need regression testing or API modernization?
5. Can we rehearse the complete upgrade and business-critical workflows on a test database?

A version upgrade is successful when the business processes remain correct—not when the login page says “20”.

11. A practical Odoo 20 readiness checklist

Before approving an upgrade project:

• inventory installed standard, custom and third-party modules;
• identify every external integration and authentication method;
• list scheduled actions, server actions and Studio automation;
• identify custom reports, controllers and frontend code;
• document critical business flows by department;
• obtain or prepare a test upgrade through the appropriate official process;
• migrate custom code against the target version;
• run role-based user acceptance testing;
• reconcile important data and financial results;
• test performance, email, payments, webhooks and scheduled jobs;
• test offline workflows if they matter to operations;
• test AI tools and permissions separately from ordinary UI workflows;
• define a production freeze and cutover plan;
• validate the production database immediately after cutover.

Frequently asked questions

When was Odoo 20 released?

Odoo’s official release-notes page lists September 2026 as the release month for Odoo 20.

What is the biggest change in Odoo 20?

There is no single feature that matters to every business. Architecturally, AI Agents, AI tools and Sources, and MCP are among the most significant changes. Operationally, offline workflows and changes across core applications may have greater impact for some companies.

Can Claude Code connect to Odoo 20?

Yes. Odoo’s current MCP documentation provides a Claude Code configuration example for connecting to an Odoo database’s /mcp endpoint using an Odoo API key.

Does Odoo 20 support RAG?

Odoo AI Agents can use indexed Sources including PDFs, web links, Documents and Knowledge articles. That enables native retrieval-grounded use cases commonly associated with RAG, although specialized requirements may still justify a custom architecture.

Were XML-RPC and JSON-RPC removed in Odoo 20?

Not entirely. Current Odoo documentation says the db service was removed in Odoo 20, while the deprecated common and object services are scheduled for removal in Odoo 22. JSON-2 is the replacement direction.

Does Odoo 20 work offline?

Odoo has demonstrated an Offline Mode for version 20 covering operations such as browsing, searching, creating, editing and synchronization. Test the exact workflows and prerequisites your implementation requires rather than assuming every feature behaves identically offline.

Will custom Odoo modules work automatically on Odoo 20?

Do not assume so. Custom modules should be reviewed and migrated against the target version, then tested with the upgraded database and the business workflows they affect.

Final takeaway

Odoo 20 is most interesting where several changes meet: AI can use company knowledge, AI can invoke controlled tools, external AI clients can connect through MCP, users can continue more workflows offline, and Odoo continues consolidating operational processes across its applications.

For an existing Odoo customer, that creates opportunity—but it also makes upgrade planning more architectural.

The right question is not:

“How quickly can we install Odoo 20?”

It is:

“Which Odoo 20 capabilities should replace our old workarounds, which customizations still create real business value, and how do we move without breaking the workflows the company depends on?”

That is the question a proper Odoo 20 readiness assessment should answer.

Planning an Odoo 20 project? Octivex Tech helps businesses with Odoo upgrades, custom module migration, integrations and AI-enabled workflows. Explore our Odoo services or contact our engineering team to discuss your environment.