Umbraco’s AI strategy

On February 3, 2026, the Umbraco Winter Keynote event was held in Odense, Denmark. This live session was headed up by the Umbraco CEO Mats Persson, their CTO Filip Bech-Larsen, and the Community Lead Emma Burstow. It gave us all a chance to discover what lies in store in 2026 and what is already available.

There were plenty of exciting announcements, but the biggest of them all was surely the comprehensive AI strategy that seems to be at odds with most of the other strategies out there. Basically, it intentionally subverts what most CMS vendors are doing.

There were many more stories to emerge from the event, and not all of them involved AI. Things like 24-hour global support, ISO 27001 certification, a 100% SLA on cloud services, and a brand-new Partner Portal are all interesting announcements. There is also the launch of the Umbraco Compose data orchestration platform and an enterprise package to get excited about.

But in this post, we’re going to be focusing on the AI announcements. In particular, the strategy, architecture, and already-available features. These aspects display a radically different approach to AI from a CMS platform.

A platform for AI

Filip Bech-Larsen made the approach very clear from the outset. Umbraco is designed to offer a platform for AI, not an AI product. This is not just about semantics and word choice, it’s a philosophy that shapes the CMS’ entire AI strategy.

But what does this mean in practice? This can be explained by looking at Umbraco’s three pillars for AI development.

1.        AI on your own terms

When you use AI with Umbraco, you choose the models, remain in control of the costs, and avoid any hidden charges. What’s more, there are no underlying functions that operate without your consent, like API calls that you haven’t personally authorised.

With this level of control, compliance conversations become much simpler. When stakeholders want to know what’s happening and why, you have all the answers because you are in charge of the AI’s function.

2. Full modularity

Another clear point from the event was that the CMS is designed to remain current, even as AI changes rapidly. New features can be developed and then installed as modules into the overarching platform.

This not only keeps the platform up-to-date, but it also means you only need to install the features you need. Umbraco supports complete customisation and a streamlined approach to platform building.

3. AI in all workflows

The Umbraco team is already well aware of the ubiquity of AI these days. And so, they aim to make it easy to embed AI into all relevant workflows.

From editing and coding tasks to asset management, the automation layer supports all aspects of enterprise operation. The keynote speakers made it clear that the automation layer sits on top of the CMS platform, allowing easy deployment of AI wherever it is needed.

AI-free core

The smart automation layer of Umbraco sits on top of the platform, it is not built into its core. Instead, the CMS is designed to be free, open, and unencumbered by additional AI functionality. This means you can choose which AI automations sit on top of this core.

The underlying architecture is separated into three tiers:

Umbraco CMS core

The core level forms the basis of the CMS platform, designed to be stable and free to use. The current LTS Version 17 is proving very popular and has already been downloaded more times than the previous three versions combined.

Umbraco.ai foundation

In this tier, you set up your configurations. These include aspects like LLM connections, user profiles, and content contexts and personas. Analytics and governance logs are also accessed from here. This level is essentially the control panel for Umbraco’s AI usage.

Feature packages

The actual AI value comes from the Feature Package level. All the prompts, co-pilots, agent capabilities, and anything built by the community are installed here. These features are modular, installable, and entirely optional. They are developed by Umbraco HQ, by Umbraco’s partner network, or by community members themselves.

This tiered approach is intended to keep things separate and avoid overcomplexity. The CMS’ base remains stable, while AI remains flexible. Meanwhile, the control tier ensures that you remain in charge.

Five, day one, AI providers

The Umbraco platform will ship with five AI connections already in place. These are:

• OpenAI, with its ChatGPT models
• Anthropic, including Claude and other variants
• Google Germini, offering a range of products in the Google AI family
• Amazon Bedrock, which is the managed AI service of AWS
• Azure AI Foundry, covering the AI models developed by Microsoft

These AI providers are not exclusive, and you can run multiple providers at the same time in the same project. In other words, you can select the right AI provider to match the specifics of the task you’re working on, so you always have the right tool for the right job.

When you install a provider, you first head to the marketplace and select a connector package. Umbraco will provide more of these connector packages over time, but the above providers are already available from launch. When new packages become available, you will be able to incorporate them straight away without a core update, thanks to the platform’s modular design.

Prompts and Co-pilot

One of the really exciting parts of the event’s keynote presentation was the live demo. This gave us a glimpse of two of the feature packages that editors and content teams will be using: Prompts and Co-pilot.

Prompts

With the Prompts package, you can attach your prepared and reusable AI prompts to any of the fields you’re using within the CMS. The demo showed us how this works across two tasks.

With alternative text generation, Prompts allows you to analyse the image and produce the descriptive alt text automatically. You can regenerate this again and again until it’s right.

With another important task, meta description writing, the platform’s AI can draft SEO copy based on content that is already on the page. This eliminates any context-switching and means you don’t need any separate tools. Prompts can deliver what you need without the additional hassle.

In principle, all of this is designed to remove the repetitive tasks that often slow teams down. The presentation made it clear that alt text and meta descriptions are just two of the available tasks. Social copy, content summaries, translation hints, and almost anything else can be generated automatically through Prompts.

Co-pilot

The Co-pilot is a more agentic AI-focused tool. The live demo showed us how Co-pilot can write a summary for an author profile and then place it in the correct field. We watched as the tool gained an understanding of the context and asked for permission before executing the task.

Another task we watched Co-pilot perform was an image search within a media library. The tool searched and identified the correct image, then set it in the correct field. Manual permission ensured the user remained in control.

Finally, we observed Co-pilot browsing the internet to locate additional information. The tool was instructed to visit the Umbraco Teams page and find fun facts about one of the team members. In practice, you can use this tool to explore any document or web asset and find the information you need rapidly.

While this is a powerful tool, the demo made it clear that it always operated on a human-in-the-loop basis. Permission is requested at every stage, so the tool is an assistant rather than an autonomous agent.

Immediate compatibility

While all the AI within the CMS’ automation layer is impressive, the Umbraco team recognise that the platform won’t be the only tool in the ecosystem. This is why they also make sure that external AI tools can integrate and interact with the platform.

Take a look at some of the integrations that are already available or coming soon.

Developer MCP — live now

The Model Context Protocol development server is already available and found in the marketplace for Umbraco 16 and 17. You’ll be able to utilise ChatGPT, Claude, Gemini, or another AI agent with Umbraco. This facilitates easy content queries, structure management, and interactions using natural language.

The Umbraco MCP template is not currently available, but it’s in the pipeline. This will enable package developers to build MCP into their own packages.

Coding Agent Skills — available through GitHub

Umbraco offers open-source skills to teach coding agents like Claude Code and Cursor how to develop according to Umbraco’s own specs. The first skill set is designed for building back-office extensions. The next one to be released will cover general implementation.

We found this to be a really important update, and here’s why. Umbraco development has a habit of moving fast, with new versions, new patterns, and new APIs emerging all the time. This means enterprise teams need great coding skills if they are to keep up while still ensuring that new aspects are implemented the right way.

This is why Filip said he believed skills would be the next big thing in AI tech for 2026. And this feature helps to ensure that Umbraco users remain ahead of the curve.

Editor MCP — coming soon

When it is released, the less feature-rich MCP server will be ideal for editors to use. They won’t get the broad access they would get from the developer MCP server, so they can carry out their content operations without fear of making broader, unexpected changes.

The interface is designed for non-technical users; they can utilise AI tools within their editorial workflow.

One of the most striking lines from the keynote, particularly for enterprises, agencies, and partners, was this:

 We don’t want to monetise AI usage.

This really is quite a statement, especially given that other CMS vendors tend to use all manner of methods to monetise their own AI offerings. Per-token charges, premium AI tiers, and usage-based pricing for AI features are all common in the industry, but Umbraco doesn’t apply any of these.

During the keynote, Umbraco stressed that their foundation package and feature packages are open source. Anyone who uses Umbraco can benefit from these features without needing to worry about service tiers.

While this is certainly a striking statement, it’s actually not that surprising. It just reflects Umbraco’s broader strategy. They are constantly focused on adoption through credible and flexible solutions. They are not focused on simply creating new revenue streams from AI consumption.

This will be a relief for enterprises, as it gives you a safer long-term platform choice. You won’t be waking up to unexpected bills, and you won’t need to update or overhaul your platform whenever you want to adjust your services.

In our view, this is truly refreshing from Umbraco. We predict that this approach will drive adoption for the platform as more and more enterprises and agencies recognise its benefits.

What’s available now and next

Here’s what is currently available and what’s just around the corner as per the event:

✓ umbraco.ai Foundation + 5 Provider Connectors — available now

✓ Prompts Package — available now

✓ Agent Package (Preview) — available now, built on Microsoft Agent Framework (preview)

✓ Developer MCP — live on the marketplace for v16 and v17

✓ Coding Agent Skills — open source on GitHub

→ Co-pilot — in development, available on the mono repo for testing

→ Editor MCP — coming soon

→ General Implementation Skills — coming after back-office extension skills

It’s clear that the developments for Umbraco are coming at a rapid pace. The foundation is already there, as well as five providers and two feature packages, all of this launched on the same day as the Winter Keynote event.

The developer MCP was already live before the event, and the skills are already on GitHub. This is really valuable, as it shows that this isn’t simply a roadmap announcement; most of the features discussed are already usable today.

Want to discuss what Umbraco’s unique AI strategy means for your enterprise?

Get in touch with our team.