Codegarden

Don’t miss Umbraco’s event of the year

Get your ticket →
Book a discovery call
Book a discovery call
AI governance, built in

AI you can defend in front of legal

Umbraco doesn’t decide how AI works in your platform, you do. Agent providers, profiles, guardrails, logging, and tests all live in one place, defined once, and applied to every AI action on top. All setup and controlled by you

Get the Umbraco AI packageLet's talk
Umbraco

Define profiles once. Apply where needed.

Profiles bundle up provider, model, temperature, system prompt, and guardrails into a single reusable configuration. Marketing on Google Gemini at 0.9. Legal on Anthropic Claude at 0.1. Anything else where it makes sense. Every prompt and agent points at a profile, so behaviour stays consistent across editors, teams, and time.

Rules that catch what the AI doesn’t

Brand policy says "no competitor names". Legal says "no unverified claims". The model has no way of knowing either, but Umbraco does. Post-generate guardrails block, redact, or flag outputs that break the rules you’ve set. Every fire is logged with full context. Brand and policy enforced at the platform, not in the prompt.

Sensitive data stops here

Pre-generate guardrails inspect every prompt before it leaves your environment. Block emails, credit card numbers, CPR, NI, SSN numbers, internal codenames, anything you can describe with a rule. The guardrail works both ways: protecting the data going out, and the content coming back, protecting you and your visitors.

testing AI.png

Quality you can test and prove

AI in Umbraco isn't a black box. You have access to refine the prompts and agents your editors use, and test them on mock data before anything reaches production: different providers, different models, different content types, side by side. What "best" means is up to you: speed, cost, consistency, creativity, tone of voice, or something else entirely.

What governance looks like in Umbraco.AI

For security, legal, and platform teams who need to roll AI out with cautious and confidence. Additional features include:

Customize Profiles

Create different AI profiles and reuse them for your prompts or Copilots. Profiles hold provider, model, instructions, behavior, and guardrails.

10 AI providers available

OpenAI, Anthropic, Google, Amazon Bedrock, Microsoft Foundry, Hugging Face, Together, Fireworks, Mistral, DeepSeek. Find the full list on Umbraco Marketplace

Logging & audit trail

Every prompt and response captured with full context

Versioning & rollback

Configuration changes are reviewable and reversible

Set up Guardrails and Rules

Inspect prompts before they leave your environment, and outputs before they reach editors.

Testing & evaluation

Compare providers, models, and content types side by side. Quality-check the AI in your backoffice.
next step

AI where you're in control

AI in your editorial flow

One click, your tone of voice by your chosen model. Check out the editor-friendly features you can benefit from in Umbraco today

Check out the official Documentation

How do you set everything up? Check out the official documentation that can help you with all the "AI in Umbraco" packages and features:

Get started today!

Everything is open source, ready for you to implement and use today. Find the Umbraco HQ AI package in the Umbraco Marketplace

Umbraco, AI, and you

Adding AI features to your Umbraco setup can really take things to the next level. Let's talk about your possibilities:

Let's talk

FAQ

An open-source foundation package, owned and maintained by Umbraco HQ. It’s where you configure AI providers, profiles, guardrails, logging, and tests — once — and apply them across every AI feature you adopt on top.

AI is moving fast. We keep the CMS stable and predictable, with AI as a modular layer on top. You choose which packages to adopt, which providers to use, and how AI fits your workflow, and you can change any of it later without a migration. The best way to stay future-proof. 

Yes. Umbraco doesn’t broker AI calls or take a cut. Bring your provider API keys, configure profiles, and Umbraco routes the right prompts to the right providers. No hidden costs. No forced models.

Rules attached to profiles, enforced before a prompt is sent (Pre-Generate) or after the response comes back (Post-Generate). Regex-based or LLM Judge–based. Every fire is logged with full context.

The editor sees a clear message. The event is recorded in the Logs view with the prompt or agent, the rule, and the full context, so admins can review, tune, and adjust.

Yes. Tests are defined once and can be re-run against new providers and models any time. Validate that existing prompts still produce acceptable results before promoting a new model to production.

Prompts and content go directly to your configured AI provider, nothing is routed through Umbraco infrastructure. The provider’s terms and region apply.

Yes. Umbraco.AI, Umbraco.AI.Prompt, Umbraco.AI.Agent, and Umbraco Copilot are all open source, owned and maintained by Umbraco HQ.

Umbraco.AI and Umbraco.AI.Prompt are on the Umbraco Marketplace now. AI Tests and Evaluations are available in Umbraco.AI now as well as Umbraco Agent and Copilot. Follow the product roadmap for current status and upcoming features.