https://discord.umbraco.com logo
Introducing Umbraco AI Search | mattbrai...
# news
r
📣 New post: Introducing Umbraco AI Search https://mattbrailsford.dev/introducing-umbraco-ai-search
w
Nice stuff @Matt Brailsford So is the underlying AI Provider/LLM only called to help store the vectors on content creation, update? Or is it called anytime a search query happens in the backoffice/frontend when performing a search?!
l
So we can also use this for site search in the 'frontend' of a site?
m
@Warren Buckley Both. You create embeddings (vectors) for all content and store those (this is what the package does) and then you create embeddings for your query and then find entries with similar vectors. This isn't anything you need to worry about though as this is all done internally and is just exposed via the Umbraco.Cms.Search API. @Luuk Peters (Proud Nerds) we have built it mostly to use in AI tools, but it's ultimately Umbraco.Cms.Search index that is exposed like any other index so yes you can use it for external search too. Bare in mind, every search query though would require an embedding created which will use tokens. You might want some caching layer if you were to expose it on the front end to cache common queries.
w
That was my concern if used on FE, could end up quite costly if used as public search on a site if you had a lot of traffic. Curious to see how much cost/token/api credits this would use/consume.
m
Wonder how well something like Ollama would do, for a frontend search
l
We currently have a custom AI frontend search and that takes some tokens, yes. But the results are great. I'm going to research if we want to use this as a replacement or as a starting point 🙂
The more Umbraco native, the better
m
@Luuk Peters (Proud Nerds) I'd love to hear your feedback on this. As I say, our primary focus is making the AI search tool more useful for copilot, but if there is anything we can do to make it more useful as a general purpose search, then that would be a win win.
l
Yeah it depends on what it can and cannot do. I think we have too many requirements at this point because it's a search for hospital sites and there are so many subagents and safeguards to prevent it from giving medical advice for example.
I want to put it into the Umbraco search abstraction anyway to make it easier to swap out in the future, but the AI search might give me some new ideas. If I have suggestions for improvement for you, I'll let you know 🙂
m
Cheers fella 🙏
42 Views