Odoo 20 moves toward enterprise level — the open-source ERP has finally broken through the ceiling

In July 2026, several things came together in the Odoo community and partner ecosystem.

One thing is that Odoo 19.4 has officially launched on GitHub and Runbot—the permission system has been rewritten from the ground up, merging ACLs and Record Rules into a single model called ir.access. This may seem like a technical change, but for medium and large enterprises, it solves an old problem: in previous versions, adding permissions for a user often inexplicably overwrote permissions from other modules. Now it's "only extend, never reduce," meaning that what is added stays added, without touching what you already had.

Another thing is that Odoo 20's Read-Replica architecture has emerged. Several official Odoo implementation partners have released technical predictions: a read-replica database architecture aimed at breaking the ceiling of 10,000 concurrent users. If this feature is released with Odoo 20 in September, Odoo will truly have the ability to compete with SAP and Oracle in enterprise-level deployment scenarios.

One more thing: From July 8 to 10, ASTA (AsiaSoft Technology) brought Odoo HK to debut at the LEAP EAST exhibition, which was held in Hong Kong for the first time, showcasing AI-driven ERP solutions on site. On the same day, Odoo held a partner recruitment event in Shenzhen, opening cooperation to IT service providers in the Greater Bay Area focusing on cross-border expansion.

Looking at these three things together, it’s not about how big the Odoo 20 version is, but rather that the entire Odoo platform is shifting its ecological niche—from an "open-source ERP for small and medium enterprises" toward an "AI-native platform capable of handling enterprise-grade workloads." How significant this step is deserves a detailed breakdown from me.

Starting from 10,000 Concurrent Users: Odoo's Enterprise-Level Abacus

Odoo's growth curve over the past decade has been impressive—from versions 17 to 19, paid users surged past 2.11 million, and revenue in the DACH market grew by 133%. But there's one statistic not many know: Odoo 17's concurrent user cap was around 2,000-3,000, version 18 optimized it to about 5,000, and it wasn't until version 19's framework layer that support for over 5,000 users began.

This ceiling is a hard limitation for medium and large enterprises. A manufacturing company with an annual revenue of 1 billion yuan may have over 3,000 concurrent online users just from workshop workers, office staff, and warehouse personnel. Adding dealer portals and supplier portals, concurrency can easily exceed 5,000. The reality is that Odoo's performance in high-concurrency scenarios has not been reassuring in the past.

10,000+ Odoo 20 Read-Replica Architecture Target Concurrent Users

Odoo 20's Read-Replica architecture is the key to solving this problem. Its approach is not complicated: split the database into "primary for writes, replicas for reads," distributing read operations across multiple replicas while concentrating write operations on the primary. This way, high-frequency but lightweight operations like querying reports and AI Agent reading data do not affect core order writing and financial posting.

Let me translate: Suppose your factory has 2000 employees using Odoo simultaneously, with 500 of them handling purchase receipts and order entry (write operations), while the other 1500 are viewing reports, checking inventory, and generating analyses (read operations). Under the traditional architecture, these 1500 read operations would slow down the database, affecting the write operations of those 500 people. Read-Replica separates these two groups, so they don't interfere with each other.

This architecture is not new—large internet companies have been doing this for a long time. But for an open-source ERP, natively supporting it within its core product, rather than having users build their own middleware to handle read-write separation, is a different matter entirely.

19.4 Permission Rewriting: Another Foundation for Enterprise-Level Development

Let’s go back to the ir.access permission model of Odoo 19.4.

I previously chatted with a friend who does Odoo implementation. He mentioned taking over a project where the client's permission configuration was so chaotic that "the same user's permissions conflicted with each other across different modules." The problem stemmed from Odoo's early permission design: ACLs (Access Control Lists) and Record Rules were two separate systems, and modifying one could inadvertently overwrite the other.

19.4 merged them into a unified model, while introducing the rule that "roles can only be expanded, not reduced." In plain language: if you grant the finance manager the permission to "view all invoices," it will not override their original permission to "view supplier information." This is a basic operation in SAP and Oracle, but in Odoo, it was only truly resolved until today.

Moreover, version 19.4 added an update that was not widely discussed in the Odoo ecosystem before but is highly practical: the Schema.org JSON-LD structured data layer. Content such as job postings, events, products, courses, and Q&A on the website now has a unified data format. Those who have worked on SEO know what this means for search rankings—but for business users, the greater significance lies in the fact that when AI Agents read Odoo website content, the data structure is clean and requires no additional parsing.

LEAP EAST + Shenzhen Recruitment: Odoo's Asian Ambition

From July 8 to 10, LEAP EAST held an exhibition at the Hong Kong Convention and Exhibition Centre that caught the attention of industry insiders. This brand originates from the LEAP conference in Saudi Arabia, which facilitates over $25 billion in investment collaborations each edition. This year, for the first time, it stepped out of the Middle East and settled in Hong Kong. Among the more than 300 exhibitors, one booth stood out particularly—ASTA AsiaSoft Technology, in collaboration with Odoo HK, set up a joint AI+ERP booth.

I looked up the content of the live demo on the ASTA official website: AI sales forecasting, intelligent warehouse management, omnichannel retail, AI CRM, and financial automation. Among these, the AI sales forecasting part uses machine learning to analyze historical data and predict market trends—this itself is not unusual; what is unusual is that Odoo official partners have already been able to package these things into a live demo solution, rather than just concepts on a PPT.

Then on July 16, Odoo held a partner recruitment event at The Langham, Shenzhen. I looked at the agenda: discussing the market landscape of Chinese enterprises going global, breaking down Odoo's partner support system, and exploring digital and intelligent solutions for cross-border expansion. The target audience was clear—IT service providers in Shenzhen specializing in foreign trade and cross-border e-commerce.

Two things are connected: LEAP EAST is Odoo's brand debut in the Asian enterprise market, and the Shenzhen recruitment is Odoo's physical move to establish a presence in the Guangdong-Hong Kong-Macao Greater Bay Area.

This also aligns with Odoo's official data. According to the data shared at Odoo Partner Day: revenue in the Indian market grew by 120%, with over 170 Gold partners. Although China's data was not disclosed separately, judging by the pace of Odoo's simultaneous presence in Hong Kong and Shenzhen, Southeast Asia + Greater China is clearly placed at the top priority for global growth.

Native MCP Server: Opening the Door to Enterprise-Grade AI

Speaking of AI, what I find most interesting about Odoo 20 is actually its native MCP server.

Odoo's R&D team revealed in early July: Enterprise users can directly connect AI assistants via the /mcp endpoint. No external Python gateway needed, no third-party middleware required—Odoo itself is the MCP server.

The user community's reaction was straightforward: "We've been waiting for this feature" and "It will make competitors tremble again."

But what concerns me more is another set of data: the MCP Enterprise Adoption Report released in early July shows that 78% of enterprise AI teams are already using the MCP protocol in production environments, 28% of Fortune 500 companies are running their own MCP servers, with monthly SDK downloads reaching 97 million, and over 9,400 public MCP servers available on the market.

Odoo chose this juncture to make MCP a native capability, timing it very precisely. AI Agent does not "access" Odoo through an API, but instead "lives" inside Odoo via MCP, using Odoo's own permission system, data format, and workflow—this is a completely different architectural choice from the past practice of building an "AI shell" on the outside.

And it's not just Odoo moving in this direction. The Agent-Ready Tools just released by Workday DevCon also use the MCP protocol. Both the open-source camp and the enterprise SaaS camp in the ERP circle have simultaneously chosen the same protocol to serve as the "USB interface" between AI and ERP. This is no coincidence.

Odoo 20 Preview: I Pick the Three Most Important

Odoo 20 is expected to be released on-site at Odoo Experience in Brussels from September 24 to 26. Based on the currently leaked roadmap information, here are three most critical points worth the attention of enterprise managers:

First, Agentic AI shifts from assistance to execution. Unlike the passive "you ask, I answer" AI in version 19, version 20's Agentic AI can autonomously monitor processes, pull data across modules, and draft purchase orders. Odoo's AI is no longer just a "smart assistant" but a "digital employee" capable of independently completing tasks within preset rules.

Second, AI Action Audit Trails. This may be even more important for medium and large enterprises than Agentic AI itself. Odoo 20 plans to establish a complete audit trail for all operations of the AI Agent — what data was accessed, what decisions were made, and what workflows were triggered. For regulated industries such as finance, healthcare, and manufacturing, AI Agents cannot be used without audit logs. This feature could be a true game-changer.

Third, Read-Replica breaks the concurrency ceiling. 10,000 concurrent users, read replica architecture, intelligent cache layer — if these three things are delivered as promised, Odoo will no longer be a tool "exclusively for small and medium-sized enterprises." For companies considering migrating from SAP or Oracle, the second half of the year is worth taking a serious look at Odoo.

In closing, I want to say: this upgrade of Odoo is not just patching up the existing framework. From the rewrite of the permission model, to the native integration of MCP, to the Read-Replica architecture—these three things are aligned in direction: enabling open-source ERP to handle enterprise-level workloads while developing AI-native capabilities.

See you at Odoo Experience in September.