Extending Genie Agents (formerly Genie Spaces): APIs, MCPs, Integrations, and Embedded Agents

Summary

Genie Agents—the evolution of Genie Spaces—extend text-to-SQL conversational analytics with the ability to answer questions on documents, connect to external systems via MCP, and reach users directly inside Microsoft Teams, Slack, Confluence, and custom web applications.
Two paths are available for bringing Genie outside the Databricks Data and AI platform: native integrations with Teams, Slack, and Microsoft Copilot Studio, and a developer-facing extensibility suite of Conversation APIs, SDKs, and embeddable iframes.
Governance and access controls for all Genie Agent interactions are tracked in a monitoring tab, ensuring that every conversational analytics session across integrations remains auditable and secure.

Extending Genie Agents (formerly Genie Spaces): APIs, MCPs, Integrations, and Embedded Agents

Genie Agents (fka Genie Spaces) delivers trusted conversational analytics—and now it can reach your users wherever your they work. With Genie Agents, users can now talk to their data directly from the tools they already use: Teams, Slack, Confluence, or custom web apps. Whether you’re using our native Copilot Studio/Foundry integrations or building with the Genie Conversation APIs/SDK, Genie can now bring natural-language analytics into everyday workflows. In this video, we’ll show how to build and deploy Genie-powered experiences across your organization: Build custom conversational apps using Genie Conversation APIs and MCP Leverage native integrations with Teams and Slack, and extend into Microsoft Power Platform and Copilot Studio Apply best practices for secure, scalable, and governed integrations
Talk By: Ambarish Dongaonkar, Sr Specialist Solutions Architect, Databricks ; Hanlin Sun, Product Manager, Databricks ;

Chapters

FAQs

What is the difference between Genie Agents and Genie Spaces?

Genie Spaces was a text-to-SQL tool that let users query data through a conversational interface; Genie Agents is the evolution of that capability. Genie Agents can answer questions on documents and files, connect to external systems via MCP, and work across tools like Slack, Teams, and custom applications—going beyond the table-and-SQL boundary that limited Genie Spaces.

How do I integrate Genie with Microsoft Teams?

Databricks provides a native Teams integration that connects Genie Agents to Teams channels so users can ask data questions without leaving their collaboration tool. The integration is configured through the Databricks Data and AI platform and supports access controls and governance tracking via a dedicated monitoring tab.

Can I build a custom app using Genie Conversation APIs?

Yes, Databricks offers Genie Conversation APIs and an SDK as part of an extensibility suite for developers who want to build custom conversational analytics experiences. These APIs allow teams to embed Genie-powered natural language queries into any web application, with governance enforced by the Databricks Data and AI platform.

What is on the roadmap for Genie Agents?

According to this video, the Genie Agents roadmap includes document analysis capabilities, third-party write actions via MCP, and scheduled tasks that allow agents to proactively surface insights. These features are planned to extend the range of tasks Genie Agents can complete on behalf of users beyond the current conversational analytics scope.

Full transcript

[00:08] Welcome everyone. We really appreciate you all being here today on the last day of Data AI Summit. Hope you've had a wonderful time up until now. My name is Hanlin. I'm joined by AD and we're super excited to have this amazing turnout as we explore integrations, APIs, and
[00:20] embedded experiences for Genie agents. So, first a quick agenda. We'll start with a level set on what exactly Genie agents are. Next, we'll cover our native integrations we have with our partner
[00:32] ecosystem. From there, we'll explore how developers can build custom experiences with our extensibility suite. And then AD will tie it all together with a live demo showing you how to make these integrations in real time. And at the
[00:44] end, we'll definitely make sure to leave plenty of time for your questions. So, I'm sure you've all heard a lot about Genie this week, and I'd like to make sure that we're all just on the same page of what exactly Genie agents are.
[00:57] So on the data bricks platform, Genie is our suite of AI features that help any user accomplish any task and answer any question. There are three main experiences. Genie one is the simplified home for business users for them to
[01:11] chat, interact and discover any data on their platform. Genie code is the AI coding and data assistant for your technical practitioners inside of the workspace. This helps with more technical things like building ETL
[01:23] pipelines, managing compute, building ML models, things like that. And finally, Genie agents are the low code agents that anyone can build to tackle specific business areas that are very curated and targeted.
[01:36] Now, as we talk about Genie agents today, I wanted to first acknowledge that Genie agents are actually an evolution of what many of you are using today, which is Genie spaces. Up until now, we've seen many customers build and deploy Genie spaces as a texttosql tool.
[01:51] Now, they've been super helpful in answering your data questions, but when users want to go beyond tables, they hit a wall. Customers wanted Genie Spaces to do more, but had to turn to other tools to accomplish this, like custom unified
[02:04] agents and and multi-agent supervisor frameworks. Your feedback has led us to evolve Genie spaces into Genie agents fully capable of all the tasks that your teams need. Soon your Genie spaces that
[02:17] you've already been using will be able to answer questions on files and documents and also connect to external apps via MCP so you can retrieve context and take actions. So just to be crystal clear, if you already are using Genie
[02:29] spaces today, there's not an additional product for you to enable or you know Genie spaces are going to be deprecated. you're just going to be able to see new capabilities in the tools you're already leveraging.
[02:41] So, as our customers started using these Genie agents, we constantly heard that agents are only useful when they're available where your users live. Customers want agents inside of their department homepages, CRM systems,
[02:54] customer success portals, and many, many other environments. To help you accomplish this, we've developed two main suites of features. First are native integrations into the most common thirdparty work applications. This
[03:06] includes things like Teams, Slack, Copilot Studio, Foundry, and Gleam. Additionally, we have a set of extensibility tools to help you build custom agents, apps, and workflows. This
[03:18] is done through our conversation APIs, our managed MCPs, and our embedded iframes. Let's walk through each of these options in a bit more detail. First, let's talk about the native integrations we have with our partner
[03:30] ecosystem. The most common use case we've heard from our customers is deploying Genie agents to where your users already talk so they can get answers and take actions the moment they come up in discussion.
[03:43] To enable this, we have a direct Slack integration in public preview today. All users have to do is simply at Genie and then accomplish their task in a group channel or directly in a DM with just
[03:56] the bot. Agent builders can keep full governance and monitoring as all of the questions submitted to your Genie agents through Slack are tracked in the Genie agent monitoring tab. To enable this
[04:08] experience, a Slack admin just has to go install the data bricks Genie Marketplace app, which you can get to from the tiny URL link above. And then, you know, to make sure that each user only sees the data they're
[04:22] entitled to, every answer generated from this integration always respects Unity Catalog permissions. So user one only sees the data that user one should see. Additionally, Slack channel managers can even configure the message v visibility.
[04:36] So let's just say for example, you're in a public channel and you don't want, you know, user one's response being visible to other users. You can make it so that all Genie responses are private to other users in that channel.
[04:49] The same story applies for Microsoft Teams. In Teams, Genie can also be accessed in group channels or directly in DMs with a simple at databicks Genie. All messages sent through Teams are also tracked in our Genie agent monitoring
[05:02] tab for full visibility and monitoring. And as expected, each user's data bricks permissions are applied to ensure data security. To enable this experience, an M365 admin just needs to go in and install the Data Bricks Genie app from
[05:15] their Microsoft marketplace. Now, some of our customers built on the Microsoft ecosystem also want to extend Genie further to their other tools. To support this, we have a direct integration with co-pilot studio. So,
[05:29] let's just say you've built some domain specific Genie agents inside of data bricks such as, you know, your finance bot or your product adoption agent. You can then expose them via MCP to Copilot Studio. From there, you have C-Pilot
[05:41] Studio agents that can be deployed with OOTH or service principles across your M365 ecosystem. So I can bring this into my M365 co-pilot. I can, you know, empower my SharePoint authors with
[05:53] context from Genie. And you know, as users are authoring emails via Outlook, they can get a quick snapshot of their account activity from Genie. I did want to mention that, you know, we, as we talked about, we have a direct Genie to
[06:05] Teams integration path. So, it's a lot easier to get started talking to Genie, but you do also have the option of building a more generalized co-pilot studio agent that is empowered with Genie and other tools and then being deployed to Teams as well. Additionally,
[06:18] if you're a developer looking to build more bespoke AI experiences, Genie agents are also available via MCP through Microsoft Foundry. So, that covers our main suite of native integrations. And now, how do you build
[06:31] more custom experiences with Genie? First up is our Genie APIs. This suite of tools lets you embed Genie agents intelligence into custom web apps, AI assistants, and workflows. So very
[06:44] concretely, these APIs lets you talk to Genie in chat mode, retrieve answers, attach user feedback, and also retrieve full conversations if you're trying to recreate a whole chat experience inside of your application.
[06:56] Additionally, authors who are looking to build these agents programmatically can run benchmarks and apply CRUD operations via API to speed up their creation process. This suite of features will continue to grow in the future as well. We'll soon
[07:09] be launching agent mode which is the deeper reasoning longer experimental process that a that Genie agents are able to apply as well as retrieving visualizations so you can display those beautiful charts that Genie is able to generate wherever your users work.
[07:22] Another piece of this is the app delegated authentication. So this way your app can manage the user authentication and translation into data bricks permissions for more simpler user access governance story. A couple best
[07:35] practices to note as you're using your APIs. First, make sure your code logic pulls for the status updates every second so that you get the most up-to-date responses from the Genie agent. And next, make sure to start new
[07:47] conversations for each user session to give Genie a clean slate into answering questions properly. To build custom agentic experiences, we also have data bricks managed MCP servers. These MCPs let you expose Genie
[08:01] as a tool to agent builders like Langraphph or AI developer agents like Cloud Code or even your general AI assistants. These servers are ready to use and managed by data bricks. And you can also pick if you want to talk to
[08:13] Genie 1, which is kind of like the unified chat experience, or if you want to talk to specific targeted Genie agents, you can pick. And finally, you know, prompts submitted via MCP are always respected by Unity
[08:25] Catalog to ensure that every agent and user only sees the data that they're supposed to. And finally, if you just want to bring Genie into your web apps, we also have embedded iframes. This enables you to do
[08:37] things like embed Genie agents into your Confluence department homepage or a custom web app showing you your task workflow as as a customer success agent. There's also governance around this. So
[08:50] data bricks admins can specify a specific allow list of web domains that Genie agents are allowed to be on. That app delegated authentication piece also helps with this story so that when users sign on to your web app, their user
[09:02] credentials can be translated into data bricks credentials as well. So before I hand it over to AD to show us some of this integrations live, I did want to call out a a few tiny URL links.
[09:15] Um so this will take you through public documentation and blog post to make some additional integrations that we have with Google chat through the APIs, you know, more detailed walkthrough in the copilot studio agents APIs, things like
[09:27] that. And with that, I'll hand it over to AD to show you how to make these integrations live. Thank you, Hanlin. How are you guys excited to see the real demo?
[09:44] >> So, I hope my demo gods are with me today. And let me just switch out some contents. The bottom one, the HTML. >> Yeah, that one. And then change. Choose this one. Choose content
[09:56] window or entire screen. Sorry. >> Yeah. >> Yeah. There you go. Okay. So first I will start with u the genie space or genie agent right. So this is my genie agent u or genie space which
[10:09] I'm going to connect with everywhere we like to see. Uh let me just make sure I have my warehouse up. It went down.
[10:23] Okay. So first thing you heard like we have the the direct integration from product with the slack and teams right. So I will start with the slack first. So what you see on the screen is a slack marketplace where we have the native datab bricks genie app available for you
[10:35] and as a slack workspace admin those can go and install the app for your organization and then as a user uh if you have the permission you can add that slack to your channel you can start DM and all of that. So let me now switch to
[10:48] my slack and show you what exactly you see. So once you have the slack admin installed the app you will see that in your app section and you see here like the app has the all the detail what can it do and next is now I have the app and
[11:01] I want to connect it to maybe my one genie agent or you might heard the genie one. So we have that configuration in the slack app directly that how you do it. So here you can see it I can decide which workspace I want to connect to and
[11:14] this is a genie one versus genie agent selection. So if I want this chatbot just focused on my GTM genie, right? Because my target is sales team. So I I
[11:26] might want to just do this in in a selected cases where I have the selected Genie agent configured. But I think I would I would say like most of the time let it go to the Genie one because Genie1 is more intelligent to route your request to the right Genie agent. So
[11:39] once you have this done here, I just go and say update and it's ready to go now. So what I did here is I have my channel uh where I have my team me and Hanelin
[11:52] working together with a genie here right so let me ask the question to genie here and I have my notes here so that I don't type too much and take too much time so
[12:06] what I'm going to ask is show me all the open opportunities I have right and then just ask the genie here
[12:19] and as you see uh Genie will go and do it stuff and as as Hanlin mentioned you're not losing anything so because I already logged in before the session so it's not asking me to authenticate otherwise it will ask me to go and authenticate with my credential so that it is following my identity to every
[12:32] place and if I go back to the Genie app for a second while it pulls the answer you will able to see that authentication as well here.
[12:51] So here when I go and select the workspace basically it will ask me to go and authenticate and based on your organization it will go through multi- uh step authentication and it will get you signed in. Going back to the sales channel.
[13:05] So it it it found the genie space and because I gave the genie one access here I have lot many genie agents to go through and make sure the right genie agent answer my question. So it it's working on it right now and as you see
[13:17] it pulled out the answers. So you can see like I have nice full detailed opportunity list available for me and with the slack app you can see like it had small filter within the slack I can look through and stuff and not only the
[13:30] data right it is showing you me the key highlights like what exactly it has. I can see more about it and it is showing me the sources. So as a user I know like I'm getting a data or the response from my trusted resources. So this makes sure
[13:43] like when you release a genie to the target audience they are trusting the data and the genie itself that the the right source is feeding the data to your genie. So now let's move on to the second
[13:55] topic. Uh we have the second native integration as a teams right and teams follow the very similar channel like a slack. So on the team's marketplace you will find the Genie app. Uh again
[14:07] administrators can go and install the app. Once you have the app installed basically it has a very similar configuration like you'll see with the agent in the slack. So if I go here so
[14:20] this is my app. You can review like you can see everything. Uh and right now I have better or admin access. So you might see more but based on your access uh granted you will see
[14:32] maybe little less than what I I show here right. So now this is my channel I created and as an example if I go and show you here manage team members and you see the app section
[14:45] here. So I created a channel or the the team channel and then I added a app here right. So this is again based on your access you will able to do it and once you have it here you can start asking question
[15:03] so this is the one question I just prepared uh to save some time like I asked how many open opportunities same question so I'm trying to show you the all the possible options and using the same genie so that you can understand that one genie agent can go to wherever your target audience is right so we just
[15:17] saw the slack now we're seeing the teams integration asking the same question and the formatting is depending on the slack and the team. So one might do a little different than other but it's the same data, same information coming out. So
[15:29] you'll see this is exactly the same information we saw and based on your ecosystem you have Microsoft teams or the slack you can pick and choose how you want to do it. I think moving on to the next. So now we
[15:43] talked about teams and slack integrations right where handling also mentioned like we have the iframe embedding right and now rather than integrating with just a communicators you can take the genu
[15:56] experience as is like full UI Q&A visualization everything and go to the app wherever your target audience is it could be confluence it could be any Jira page it could be in anywhere in your organization any third party app or it
[16:08] could be datab bricks app so how to do it right so now I will go back to my genie space and pick up my genie space and on the share basically if you click here I hope
[16:20] everybody sees it well. So this is where you do the access control around Genie right and now you have another option here embed the agent and when I click there I can see my embedding code. So as a developer or application author or
[16:34] anything you don't need too much to go figure out how I embed this. So datab bricks genie will give you the code like just take this and put it in your application have the right rendering and UI space for it and that's it. So let's do it together in a live session. So
[16:47] what I'm going to do I'm going to take this because I'm going to add it to the confluence. So I don't need everything. Conference do have the I iframe. So I go edit this page.
[17:06] I take the iframe here. I give the URL. And one thing I want to just make sure is I know we need a height 600 and width 100%. That just the UI aspect.
[17:18] Oops.
[17:33] So once I have it, you see it's it's bringing up here. And just to save some time, I have just the copy of the page where I did the embedding. So you will see the same confluence now have more
[17:45] capability to do than it was doing before. Right? So now you see this is the same opportunity genie which I have here. So if I have the PMs looking to their daily checklist, daily routine of the stuff they do and they have some
[17:57] questions they can ask the right away in the same UI. So as a user you are saving your time you are being more productive and you are just taking a help of Genie agent to do what you need or go and ask some more deep dive questions rather
[18:09] than going to different apps or anywhere else. You can do it in the same place and again it is going to follow the unity catalog governance model the same authentication and stuff because I have the same thing logged in it's not asking me again so this is the uh second type
[18:23] of or the third type of integration we saw the slack teams now this is embedding again this is very um restricted place where we have the the wikip pages and stuff but what about my external apps right or I have my third
[18:36] party app where I want to take it to maybe outside of datab bricks ecosystem and still show the same genie embedding experience we can do that so the way I'm going to show that is so um this is
[18:48] another genie space for that and what I have here is this is again separate genie separate account and this is my app which is running in Azure like a public facing app right so this is fully
[19:00] public app not at all restricted to anything else and it I can still able to embed my genie here it has a right real estate available on the page and you can see I can ask the questions Let me just copy my question
[19:26] Okay, give me a second. So, let me just go back to that UI.
[19:39] So this is um just recorded this is a no taxi data set. So I'm not like focused on what data I want to show. I want to show you what you can do with a genie agents right. So I have my app which shows what KPIs or stuff my want app want to show
[19:52] and on the side I have the Q&A interface or a genie intelligence agent available for you to directly go and ask any question you like on that data. So this is easier to do like I showed you the
[20:04] iframe right. So how I'm doing it with the external app? Let's look at the code while it pulls the answer. So this is my web code like this is very simple page but you can have more advanced code and stuff doing on your web page and you
[20:17] just need the simple iframe embedding there. So you see this is a code which is getting you the whole agentic experience in your web app now. And this is again going outside a databicks app and doing this so that you're not
[20:29] restricted just to take a genie and stay in a database ecosystem. you can take it and put it in your apps again which might be governed through your VPN and stuff. So it's not like going on public internet for that matter because you don't want to break out your corporate
[20:42] network and go out but if you really want that's coming up as he highlighted the app delegated authentication and stuff coming so if you really need that it it can be done and on this on that topic we do have the API support right
[20:55] so this is embedding experience but API can take you anywhere you like so moving on to the demo sequence next is a Google chat so now you you've
[21:07] seen the stuff so far with the teams, the slack which covers most of the communicator ecosystem but now there are lot many GCP customer who use the Google chat or they have the the full Google
[21:20] ecosystem where they want to stay in. So Genie can now integrate with the Google chat as well and again similar concept like you can go ask the question in the Genie Google chat and stuff. So integration is slightly different. So
[21:33] the way we can do right now is you can use like I'm for the demo purpose or giving you the conceptual design how you can do it. I'm using a Google app scripts. I know that may not be the your choice of compute and stuff in the
[21:45] Google but you can choose the way you like. Basically this is a middle layer who can talk to your genie agent and the Google chat APIs. So the way to work with Google is Google chat has the API where you can go and connect your middle
[21:58] layer and middle layer serves the genie agent and the Google chat communication and just to show you that like this is my agent. So I configured the workspace detail you saw up um in the slide in the
[22:10] content and this is the way you can do the polling all of this is using a genie APIs. So Genie APIs when I say that it's a genie conversation APIs which are been there for almost two years now uh which
[22:22] are really matured those APIs can allow you to give the feedback loop and everything and as you just saw in earlier slides the the visualization support is also coming to the APIs so that it will be fully rich in the
[22:34] features you need to support just going deep into how we do it in the Google chat today. So this is my uh Google chat API and on the configuration side on that API basically I created a app. This is my DGI app and this is the
[22:49] logo for that and basically I can define at the API level how want to do it like this is my app script. This is my deployment for that and all of these commands I want to get uh invoke with the genie and right now for my demo like
[23:03] I opened this for all the organization in this. So let's go and see the demo here. Um let me ask a different question.
[23:16] So pick a genini which is longest trip. So you can see earlier question here I asked for a shortest trip. So now I ask
[23:29] for a longest trip. And as I said before you might seen the different UI is rendered on the Google chat versus Slack. And that's the capability the the communicator has right and Google chat may not be that fancy like a teams or
[23:41] the slack but you can do better than this if you really have more UI aspect in mind you can have that appcript render it very differently you can look at like how the analysis can be shown how the trip data can be shown and you
[23:53] can get creative with it but the purpose of my session is like show you what is the art of possible is so it's not limited to just API just embedding you can do everything and as you
[24:06] So if the answer is good, I can say it's a good response and then the feedback gets recorded to the genie and uh that's one part of it. So let's let's look at now the another integration. So
[24:20] before I show that, okay, last page. So now we saw um we we talked to the genie in the datab bricks UI. We have
[24:32] the genie one. We have all the integration with the Slack, Teams, Google chat, we did embedding, we did in internal app, external app, everything, right? What if you could go and talk to the data directly, right? Like when we
[24:44] say that we we we mean it. So we're not just typing the stuff on the chat or mobile and anywhere. You can really go and talk to the agent now. And the demo I'm going to show you is with Alexa.
[24:57] So I can go and talk to Alexa directly and tell go pull my data set or ask the question to Alexa directly. So let's see that in a action. So I have the video recorded just for the demo.
[25:13] >> Alexa start. >> Oops. Let me see. Give me a second.
[25:29] Alexa, start databicks genie. >> Here's Genie. Say Alexa exit to get back to Alexa. Welcome to Data Bricks Genie. You can ask me questions about your data. For example, you can say what were our sales last quarter? What would you
[25:41] like to know? >> Alexa, can you tell me how many trips are recorded in the data?
[25:59] There are a total of 21,942 trips recorded in the data. >> Thank you, Alexa. >> So, what you just saw is like Alexa doing the the cloud or the its own processing. That's the Alexa skill. And
[26:13] let me pull up the code to show you how we can make it happen. So again it's the same integration using APIs. So uh on the API side we we I built out a genius skill.
[26:34] I think that that concludes the demo. So basically um my session got logged out so it will take few minutes for me to log back in into it. But it's the same integration like using APIs. I showed the earlier the datab bricks Google chat and the genie integration. The same
[26:47] interface works well with it's a genie Alexa skill set and I I use the AWS lambda there to kind of integrate it with a genie agent. So my communication part there is a lambda in AWS cloud
[27:00] which can learns the genie skill and get invoked based on a gen Alexa commands and talk to the genie and gets the response back and translate it to the genie genie to Alexa prompts. Thank you.
[27:26] are able to talk. Uh, you know, whether that be literally through Alexa or in Slack and Teams, you know, your custom web apps, your custom web pages, Confluence, wherever your users live, we want to be able to help you deploy your
[27:38] agents to. Uh, just to give a quick road map as well to make sure you guys are all caught up on what's coming with Genie Agents as well. As we talked about, there will be document analysis capabilities and all of this will be available through our extensibility suite as well. So you can analyze your
[27:51] documents alongside your tables for insights. Also being able to connect to thirdparty tools. So you're going to be able to read context from these SharePoint, Google Drive, you know, additional resources and even take actions in them, right? So can I file a
[28:03] Jurro ticket based on the analysis where I detected some anomalies? Can I uh, you know, create a Google doc that, you know, analyzes and and summarizes certain processes within my company? And those are, you know, a part of our
[28:15] our bigger road map. And there's always kind of key three key themes we're trying to push on, right? first is always trying to expand the capabilities of what the agent can do. So in addition to MCPs and file support, we're looking to help with scheduled tasks. So can I
[28:28] schedule a certain type of analysis to be done every Monday morning for my you know account teams to to take a look at uh custom MCP tools as well. So if you have custom experiences or or systems of record that you want to plug into Genie,
[28:40] that should also be possible soon. Uh and next is making sure that you have as people building agents all the toolkits that you need to build them very effectively, right? Um so there's a lot of new integrations with Genie code so
[28:52] you can actually have an agent build your agents. Um and and really help you through the debugging and setup process. Uh and finally, we want to make sure that nontechnical users, the business folks always have a very clear experience on how to leverage agents
[29:05] effectively as well. And so this is including things like chat sharing so you can circulate insights and get verification from your data professionals also as we talked about right when you're a business user accessing Genie from external sources we
[29:18] want to be able to allow you to leverage those visualizations via APIs the longer reasoning mode and agent mode and having a seamless user governance experience with app delegated authentication.
[29:30] Awesome. So that concludes our slides and demo portion of this. We've left about 9 minutes to go through Q&A. So, really appreciate you folks taking some time on the last day of Data AI Summit. And uh yeah, definitely let us know if you have any questions.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.