Or should I say Team x AI x Copilot Studio. Just recently I presented this topic at CollabDays Poland @ Microsoft Warsaw. This is the same topic I have already given a few times, but this session is always different since Azure OpenAI Studio and Copilot Studio are evolving at a very rapid pace! When I first did this session at Scottish Summit, adding your own data was new and then at Baltic Summit at the end of September 2023 functions were new. The last time I did this one was a February at Cloud Technology Townhall Tallinn. This time Power Virtual Agents was already renamed Copilot Studio and it had gotten a lot of new features there, which I incorporated onto my session.
At CollabDays Poland the audience experienced an updated version of this session as I have added new demos and information to it.
Let’s look at some of what’s new!
Assistants API in preview New API-versions Add your data updates See you at the future events!
Assistants API in preview
This is a very new feature on Azure OpenAI Services. Basically, it allows the creation of persistent chat sessions with Azure OpenAI, and you can introduce up to 128 functions to it, files and also code interpreter which is the really cool feature. Basically, it can generate and run python code as needed. What else? Well, the persistence – it is possible to continue the conversation later and unlike with traditional GPT-model, you don’t need to carry the history in with calls. This is easy to try out with Azure OpenAI Studio , check out Getting started from the Learn. The learn describes assistants is a very concise way:
“Azure OpenAI Assistants (Preview) allows you to create AI assistants tailored to your needs through custom instructions and augmented by advanced tools like code interpreter, and custom functions.”
This is a topic that I will be doing sessions in the future, I expect this feature to develop in a rapid phase.
New API-versions
Latest Azure OpenAI API-versions are now generally available version REST API 2024-02-01 and the latest preview version is 2024-03-01-preview . Older API versions will be deprecated as time goes on, so it is a good idea to start updating your solutions to use newer API versions.
Add your data updates
If you have read my earlier Azure OpenAI Services blog posts, get ready for some update. You notice there are differences in the API, the biggest of them being that certain items have been renamed in both call and return JSONs.
In Add your data you have to, basically, include the change from Azure Cognitive Search to AI Search. For example, type is now “azure_search” , DataSources are “data_sources”, and QueryType is “query_type”. You can see the pattern very soon there. In add your data if you use vector index you need to add embedding API endpoint and key to the call as well.
The return JSON has been changed as well, be prepared to re-evaluate it and do changes accordingly. When I updated my Power Automate flows, it was quite easy when I got the output JSON and used that as schema example in Parse JSON.
There are also other updates, such as new data sources
RAG ( Retrieval Augmented Generation) is one of key features what makes GPT-Ais useful.
You can also use these in an easy way at Copilot Studio – just drop in your files to Generative AI and you have a knowledge base bot running in minutes!
Just note that you need to maintain these documents manually. But it is easy and you can teach this to people responsible of that content.
You can also use Generative Answers in Copilot Studio to add data sources where the information is aggregated from.
So, let’s add one more minute when you add up to four SharePoint sites to the source and then your bot has access to the information that is updated automatically.
These two methods (adding documents and urls) are not new, but I have seen them getting better recently than they were 6 months ago.
Then there is one more addition for these data sources, that is just a few months old: Adding Azure OpenAI Service Data Source.
You can do this by going to the old Azure OpenAI Studio and defining your RAG (upload pictures, add existing Azure AI Search indexes and so on) and when you have defined and tested that it works with your data, you can deploy it to Copilot Studio.
The deploy creates a new copilot for you, and automatically adds Azure OpenAI connection and properties. If you add this manually to existing copilot, I suggest you deploy a new with this and then check out properties from that to the new one. This one is very tricky that you need to use specific APIs and it doesn’t work with all models – GPT 3.5 Turbo seems to work well and using API 2023-06-01-preview with it is a setup that works in my environment.
See you at the future events!
These are just a few highlights of what is new in Azure OpenAI Service and in Copilot Studio. I have been working (and demoing) on how to extend Copilot for Microsoft 365 with Copilot Studio, but I think that is still work in progress – but I recommend testing it already, since it is getting better every month. Just like Dynamic Chaining in Copilot Studio is already working a lot better than it was two months ago.
I am not writing about AI enhanced processes (with Power Automate) on this post but just mention that Azure OpenAI Services GPT-4 Turbo with Vision is fantastic on analyzing images.
The next time I present this session, I might split this to two separate sessions – one with more overview style giving the big picture and another one that dives onto specifics on one or two scenarios. There is so much content, that I could do this as a full way workshop with attendees.
Let’s see what the future holds – but stay tuned for more updates and information about AI, Copilot, Teams, and Copilot Studio!
Before that – let’s meet at the Microsoft Community Conference 365 at Orlando, Florida at the end of April where I present three sessions: one solo and two with my good friend Adam!
And if you still don’t have a ticket to #M365Con, you can get $100 off from the price with my code NOPANEN100
