Umbraco.AI Prompts

Hi there!

I am trying to test the new Umbraco.AI features
and managed to get the Copilot chat running
as well as the Prompts
though it doesn’t seem to have context of the content.

the agent has all the permissions needed.
but asking it to fill a value for example doesn’t do much and I can’t save the page after that.

and in the case of the prompts
I have a prompt that should generate a Seo title
but it keep responding with something similar to
”Certainly! Please provide the main topic or content you’d like the SEO title to focus on, and I’ll generate a short, SEO-friendly title for you.”
or “It appears that I was unable to retrieve the current page context directly. However, based on the information I have, I will generate a concise SEO title related to the content about Umbraco AI….”

Model: GTP 4.1 nano

am I missing some configurations here?

any ideas?
please and thank you.

For your prompts, you either need to have the “Include Entity Context” checkbox checked for the LLM to understand the current content context

Or you’d have to include {{ variable }} entries in your prompt to reference properties of your content model.

From an agents perspective, we probably don’t have the tools setup yet, hence why it’s still Alpha release. We only released the permissions system yesterday and are still working on this actively. There are only a handful of simple tools currently implement (enough to demo possibilities for the winter keynote).

hey there leo,

following this article/blog

we also tried to setup the AI, but we got into an issue like yours, Nathaniel and i already reported this and it appears to be on their roadmap

i wonder how they did it ?

Hi Matt

thank you for the quick response

on the prompt, you are correct, I didn’t add the property needed. this works as expected now.
and thanks for the clarification about the agent.
this is a good start for AI integration and I am looking forward for what comes next

Thanks everyone for the clarification.

I initially thought the AI would automatically understand the current page content, but now it makes sense that I need to either enable Include Entity Context or explicitly pass the required properties using {{ }} variables in the prompt.

As for the Copilot agent, I understand it’s still in Alpha and the editing capabilities are limited at the moment. Good to know this is already on the roadmap.

Appreciate the help!