Name: Couchbase Roadmap Preview
Uploaded: 2025-11-21T18:01:06.924Z
Duration: 1 h 45 min 8 s
Description: Couchbase Roadmap Preview

Transcript for "Couchbase Roadmap Preview": Hello, everyone. Thank you for joining us. Today, we're gonna give you a brief look into, Couchbase's product road map. Give you a little preview of, some exciting things that that are gonna be coming in the next, you know, few months to twelve months. We'll give you that give you that range. I'm gonna be presenting today. I'm Tim Rotek on the product marketing team. I'm here with my boss, Jeff Morris, who runs the team, and, Mark Gamble and Matt Groves are also be joining to cover, various components of this presentation. So Jeff, why don't you kick us off with the first, with the agenda and, the first section? Very good. Thanks, Tim. So, welcome everybody. What we're gonna walk through today is, what the new, in many cases, the release defining feature set for all of the different product areas of the Couchbase platform. So we'll start out with Couchbase Enterprise, and that's the traditional server. We have just introduced version eight dot zero, and we'll go through what, the key capabilities, are included there. We'll do a review of Couchbase Capella, our cloud database as a service. We'll go through the new capabilities in, Couchbase mobile and our our overall mobile offerings. We'll talk about enterprise analytics and analytic, Capella analytics. We'll talk more about the AI services, which are also, rolling out it, right now. And then finally, we'll wrap it with, what the developers can expect with regard to, new toolings, SDKs, and the overall ecosystem in, within which we are operating. So let's jump in, and, we'll talk about Couchbase Enterprise and what the new capabilities, for that are. And so when you really look at the way that the Couchbase database platform has evolved over time, there's been many, many significant, capabilities that we put into the product to evolve it into, what it is today. Now if you look back at the twenty two thousand eleven, era, we were really the first multimodel database to come onto the market. That's when, the folks curating Memcached from Membase, met up with the, curators of CouchDB from CouchOne. And the, combination of those two teams is what bore out Couchbase itself. It was a key value database that, had a really, groundbreaking capability in in terms of supporting active active clustering. So any cluster node can, receive a write as well as offer reads to their applications. That's all facilitated, of course, through the cluster map, something that even continues to evolve today. We added new, functionality, new engines or services inside the database for SQL query, for full text search, for, eventing and streaming, as as the platform grew. And then on the operational side, we added things like being able to tune the number of services and and and the resources that services are offered within a cluster. And then you can now be able to move those around, much more easily in our most recent release. We have cross data center replication that is industry leading. We introduced a breakthrough in JSON, transactionality for JSON documents. We hold many patents for that right now. We added time series data in in the, seven dot x, time frame. We added, natural language query capabilities and Capella IQ. We added graph traversal capabilities to navigate hierarchies and whatnot. And as we look into what we've got, coming right now or or we just introduced, is billion scale vector search. And all of these capabilities are are steered towards supporting our the critical applications that our customers are deploying. Amadeus, for example, is supporting a system that's operating at 50,000,000 requests per second. That's really that's a lot of requests. We have airlines that are running their business on, on Couchbase. And, you know, AT and T supporting us with, you know, a 130 or more microservices as part of their overall deployment. So Couchbase Enterprise, that foundation and what eight dot o now offers is really, you know, quite extensive and and and quite popular in the market. So our overall vision as we look at our intention for deploying the, the the platform is to make this a developer database platform for all of these critical applications. It needs things like this, multimodal or multimodal, multipurpose capability so that you can support a variety of different workloads all from the same, technology. It's obviously, it has to support very low latency and very high performance. And, traditionally, that's what customers have purchased us for. The number one reason they they they continue to tell us why they, select Couchbase for their critical applications is performance. Our ability to cash cape cash things, our ability to support low latency, operations and transactions, or our ability to operate in a distributed manner, given that one platform, operating everywhere kind of, capability. And finally, the developer first experience. We know that, our customers, their architects are leaning on their developers to build these, mission these, mission critical systems, in a highly innovative way and do so, such that they can be incredibly productive. And so things like SQL plus plus, our query language, help, support in that endeavor. But you'll also notice that the platform itself as we've defined it has these four pillars of capabilities. The operational services, of Couchbase Enterprise, the mobile and edge services, of the, of our mobile extensions for that, The analytics services, which operate as a, an independent column store and, high performance, compute engine. And then AI services, the the extension of the entire platform to support, new development techs techniques like, delivering rag workflows or, you know, building agentic systems. AI services supports all of that. So when we look backwards at the the the way in which, the, Couchbase enterprise has evolved from, you know, in July 2021 when we introduced scopes and collections and transactionality within SQL plus plus. Scopes and collections, of course, are a way to mimic, building a a a structured schema within an unstructured database, a a JSON data store like ourselves. We introduced Magma as a new storage container in seven dot one, much more, high performant, highly performant, key value database living underneath Couchbase that supports the, the modern style of key value database. It's a log structured merge style database that, self cleans. It it it cleans up after itself and keeps a an ongoing record of all the operations you're you're supporting. We added, change data capture and change history capabilities in seven two. We added vector search in, seven six. That was operating inside the store the search engine of of Couchbase. I'll get to that in a minute because we'll, we we've added more capabilities in the areas of vector search coming up. And now with Couchbase eight, as I just said, we've added a ton of new features, especially related, to supporting, AI powered applications themselves. So for developers, if you're building these AI enhanced applications, you get some great, great features here. We're we're we're we've introduced a billion scale. We call it the hyperscale index. The billion scale vector index with that delivers extremely high performance, at scale and because it operates within the Couchbase, environment, also delivers extraordinary TCO for the amount of work we're able to do and the amount of accuracy we're able able to deliver when you're, performing vector searches, at an extremely affordable price. We added, new extensions to SQL plus plus, I'll show you in a second, for this vector search functionality. And we're integrating with a number of different AI frameworks, including the enterprise AI framework from NVIDIA. On the developer productivity side, we also embraced AI. So we added natural language, interaction capabilities to SQL plus plus. So you can write phrases and ask for things, or speak things even to, and and, the database itself will rewrite that that as a SQL plus plus query. So we, we improved our query statistics gathering and reporting capabilities so that now you can see what's running slowly and what's not in a much more manageable way. And we added some cool things in, like, full text search. You can now have a synonyms file that you, would would wanna search for. So if you're looking for, you know, synonyms to the color red, you might say pink or something like that, or is a good example as a a synonym for, for that that phrase. And then we made query troubleshooting significantly easier with this, workload snapshot with, summary reports of those collected, statistics and, you know, even including, you know, time period kinds of comparisons. So let's get into the vector search capabilities because this is really what the, release defining feature set, for this environment is. So we we've always had or we've had for since 07/06, our search vector index. And that's great for when you're, trying to incorporate, and use vectors in a larger search engine based query. So you might be looking for geographic coordinates. You might have scalar parameters that are a part of that query. You might have, search phrase text search parameters as part of that query, and also wanna collect vectors at the same time. So this gives you the broadest, way to use our our vector indexing and and vector storage capabilities is using the search vector index. So, powered by the search engine, it gives you a massive amount of utility about, you know, where and what you want to be gathering vectors for. But then we, as as we built out version eight, we added two new index based, vector indexing capabilities. When I say index based, I mean, this runs within the indexing engine or indexing service of Couchbase enterprise. And so the composite vector index is great for when you control all the the parameters that are in the prompt, and you know kind of exactly what vectors you wanna be fetching within that prompt or or alongside that prompt, in order to provide better context. So you can run a pre filtered query against that, you know, against that information to just gather the specific vectors that you want. So, you know, really great for when when as I said, your developers know exactly what they're looking for. And if they don't, and you can't anticipate what's going to be asked of your, of your large language model, and you have to perhaps you have a huge corpus of information that you want to vectorize and supply as context when you're conversing with that large language model, you would use the hyperscale vector index. And this really is the breakthrough of Couchbase eight. It it's, it will deliver best in class performance. Performance in vector land is measured in both how many queries per second can you do. So you're you know, how much work will you perform. Also, you know, also against how accurate your vector retrieval happens to be. So you can adjust and fine tune for both, for accuracy, and that might affect your query performance, you know, as you really dial up your specificity. But it's a great, great, engine to be using when you might not know what your consumers or what your, what is gonna be asked of the large language model. So you wanna provide the broadest set of vectors for, any kind of inquiry against that large language model, and supply the correct context back to it. So hyperscale vector index is really a one of a kind kind of, technology. And what we what we've done is within, vector creation, we've, you know, given you a number of new commands inside of SQL plus plus. So you can say create a vector index for your your particular, corpus of information. You specify on, you know, on what type of, data objects on movies, you might you know, or the overview vector. You you might declare when, what you're trying to index here. And then you can say the other fields alongside that you want to include in this index. You can partition it. Right? So this is a distributed index. So it will scale infinitely. 1,000,000,000 is just a random number to us. What we could scale to to four ten, and you'll see us, you know, continue that, that kind of work in the future. So for you you have a, the attributes that we've added to, for vector creation. You you index on a specific column. You can then add, optional fields to include in that index. You can partition it as I just mentioned. And then there's some other spec specifics that you're, you wanna use or specify, like, what's the dimensionality of those vectors? What is the, quantization or the description of that, you know, and, the the particular, vector query you want your or or, vector algorithm you wanna use for that particular query. And then you can specify the similarity, formula you wanna use. So in this case, it's cosine. You can say, you know, what an optional, sample set of vectors is. So it can say, you know, it looks like this. And you can say, you know, what the number of partitions that, you know, replica might live on, to go and search. So those are brand new parameters that are built into the vector index creation exercises here. And then when you're querying this, when you're searching for the vectors themselves after you built them, or you define the index, then you build them, and and, selecting on them, you've got your, you know, overall covered fields. So, you know, what you're trying to look for. So look for the, IMDB rating from that, movies collection, where your director is Christopher Nolan, and then, you know, order it by all of his films. And then you give a approximate vector index, you know, function for gritty psychological thrillers about dreams. Right? So, you know, looking for, those particular movies from, from mister Nolan, there's a handful. And so your proctor, your approximate vector index, you know, will take three, five, or so parameters, and then you can have it, you know, push down to however many, results you want retrieved, for that, particular search. So vector indexing, hyper scale vector index brand new, the things we wanna show off about it or describe about it is and prove that it scales like crazy, is we have also written up a a brand new, report, that demonstrates a, vector indexing benchmark, where we used an industry standard or, you know, a vendor standard, the benchmark that's not ours called the VectorDB bench. And we compared one of our closest competitors to, how they do in a, in a 100,000,000 vector kind of activity, and then also in a billion vector scale activity. And the results are pretty extraordinary. So when you you look at the, the 100,000,000 test, that's the, diagram here on the left. And, you can see that we're running extremely fast about what, five times faster than, our competitors capable of doing at the same quality of service or similar, recall recall accuracy ratings. So 22,800 queries per second versus, 4,600 for, for MongoDB. And then when you dial up the accuracy rating, interesting things happen here. So we're still operating at, you know, more than, like, a 90, you know, 92, 93%, accurate recall accuracy. We're, you know, more than double what Mongo is capable of doing. And then when you even go further, we we're still, you know, at 25%, more operations per second at a much, much higher accuracy rate. So and that's just with the normal data set that everybody uses. When you when we start to go on the right hand side into the billion vector realm, the results are extraordinary. So here, we're still at, you know, at at, at at at light light accuracy readings, we're still in the nearly 20,000, queries per second range when, of course, Mongo is in the single digits, for for accurate you know, for their queries per second. And then as we continue to push the accuracy, lever and, and dial dial it in for, you know, specificity here, we can still do hundreds of queries at very high accuracy rates, and MongoDB can do two per second at something ridiculous. Like, their latency on this is something, you know, in the order of, like, forty seconds per query. So really, really, extraordinary, results coming out of this. So the way we're looking at how we talk about our vector search capabilities is not only it's it's it's sort of the multiplicative effect of everything we're doing here. Not only is it providing industry leading performance with great throughput, you know, massive scale, three you know, a a a high degree of optionality for the types of vectors you wanna be using, but it's delivered through the best user experience, the best developer experience. It operates in SQL native syntax. It's, you know, it's easy to deal with, and and easy to integrate with, or or use against different, language large language model frameworks. You still, you know, can operate with, you know, your zero ETL for JSON kind of principles. So if ingestion of other external data is important here, we can do so. And all of this auto scales. Right? So your your operational activities around this are really best in class. And that all of both of those things end up leading to lowest cost of operations. So the design here is just to help you support all the vector search kind of workloads that you're going to be deploying, but also not break the bank. So and and when we start to talk about our AI services, we'll talk about the benefits of having a agent catalog and having semantic caching also supporting that TCO, argument. So you're not burning, large language model tokens, at an extraordinary rate. You're you're you're, conserving them. So, you know, and still getting the kinds of, throughput that you want out of your, your your rag system or your, your vector search activities. So we're really excited about that vector search stuff and but that's not the only thing we put into 8.o. So the next thing we did can can talk quickly about is, our automate our query, repository. So it, it automatically gathers your query statistics and puts them in a collection inside of Couchbase so that you can query and retrieve them and see what kind of what is happening with your query workload over time at, you know, under different loads and have a much, much easier time, managing the operation of, of all of your queries within, within the the Couchbase cluster. We've also added a new reporting tool for this, so you can generate reports, directly from, this information. And everything's aggregated so you can do time based comparisons. Like, on last Friday, it ran this fast, and this Friday, it runs this fast. Where's the where's the problem? You know, help you identify where the bottleneck is. So, you know, as as you as we deploy this, you know, comparing those workloads, you can take different time period statistics. That's that's what we're demonstrating here. So you can specify the overall time periods in which, all of these operations are running, kinda cool. And so, you know, those are our developer centric sets of features. And then next is what we did for your operations, people in running 8.o and managing it. So we've added a number of security capabilities, more more specifically encryption at rest. But that's, you know, not only do we do we just add, encryption at rest, we added key management functionality so you can determine how, you know, how and which keys are, encrypting that information. We made managing we simplified manage management in, within the the environment. So, we aggregate, your SDK metrics inside the database on the server side. So you can see what's happening with, client side activity, but it's stored back on the, in the cluster. We're allowing you to, add or remove or move around your different, Couch based services, like, like indexing or query or key that or or eventing. You can move those around without having to introduce a brand new node to the cluster. And so you can redeploy them to to map to the performance requirements of your application much more easily. We've made, failover and uptime, you know, improvements like, ephemeral buckets. If you, you know, overrun your memory or the if the bucket disappears, we'll automatically fail that over. And same holds true for when we run out of disk space. We'll fail over, much more gracefully and, recover from that. So some great great things for your, your SRE who's, you know, who's managing, your Couchbase cluster. So encryption at rest, it's it's native built into the platform. It'll do, file level encryption, and, it supports the data service right now or all your specific, Couchbase data that's in the Magma database. And then, eventually, we will add on to that to support your indexes and data and other services at the same, you know, at the same time. So, it's a phased approach for how we are deploying this overall to make sure that, you know, this is operating properly. And so you get your configuration data, the data service data, and your audits and and log files are what get encrypted, right now. And then key management, you know, much more much easy easier way to manage your encryption keys here. We'll, we we took a, you know, obviously, a multi key approach. So the, the encryption key, you know, in your KMS can, you know, encrypts the data using, your your specific data encryption keys. You could rotate these. We can automatically generate a key as, you know, as you need for perhaps each, v bucket. And when you rotate them, you know, you're only re encrypting the small you know, the specific, encryption keys that you're, you know, you're focusing on. So you don't have to you know, it doesn't take a lot of time to, to re encrypt or or or change your keys out. And that application, you know, service level telemetry that we're talking about, that gathering data from the clients and storing that on the server. So your SDK client here, you know, will push its its operating statistics like timeouts, like, you know, you know, operation cancellations or or latencies for the key value service, the, the query service, full text search analytics, etcetera. This is aggregated across all the clients so you can, you know, manage, high degrees of activity and explosion of activity. Server side, you can, you know, initiate that, collection. So the the server asks for it, so you're not, you know, overwhelming, the clients themselves if they're under heavy load. So and then this, you know, metrics monitoring capability, you can attach this to Prometheus and, you know, you can see what's happening, you know, in your client behavior, from with within Prometheus. And, you know, it, works as a standard log collection for, you know, our support and our engineering staff as well. So your upgrade upgrade path to getting to 8 dot o, fairly straightforward. If you're on the versions of, Couchbase seven, you can up upgrade directly. And if you're on older versions of Couchbase, you need to get to a version of version seven, and then you can upgrade directly to, to eight dot o. So key considerations as we look ahead, and and start to consider what we're going to do or or, you know, we've been working on. The next major customer asked capabilities encryption at rest. And so, you know, across all the services, so we're we're taking care of that. We're gonna add, all support for for JWT. On the disaster recovery side, inside of eight dot o is a preview of point in time recovery. So this is not officially supported, but it's in there. And then in our eight dot next release, we will, make that generally available, but we wanna start to introduce it so we can get your feedback on this as as we go along. We're supporting snapshot based backups and snapshot based restore. And, autonomous operator will also support, resilient rolling upgrade. So as you wanna be, you know, rolling out via Kubernetes and and using CAO in that regard, recovery is much, much, ease will become much easier there. Query plan, stabilization. What that means is we're making the, the the, the your query plan as we lay it out, won't introduce weird things that you might not expect. High, you know, higher degrees of detail on index scans, XTCR goes, cloud native. So that too will you know, it gets rolled into CIO and other other features. And then we'll continuously improve performance of, both vector search, performance of other services. We're gonna introduce a file based rebalance capability, so your rebalance timings will drop dramatically dramatically, and will support instant scaling of, within Capella. So let me show you what some of these things look like. Like, point in time recovery, you could specify the timing interval that you wanna, is is an acceptable loss. So you can, you know, marry that down to, you know, seconds if you like. This is, I think, showing about, like, a a five minute, five, six minute increments here, six minute increments, where it's doing a a full incremental backup and then the point in time, change log backups as it's going along. That stabilization of the query plan, you have some sort of, you know, change that, that messes it up. The optimizer optimizer will recalculate it and your performance might, might change unpredictably. So we're gonna make sure that we lock these execution plans for, you know, a consistent performance and then make every you know, ensure that everything is predictable as you you know, if if, you know, as the the so these these change surprises don't surprise you anymore. That file based rebalancing, like I said, that could dramatic change in, the time it takes to, move data around a node and, and rebalance your cluster. So you could what could take, you know, hours is gonna take end up taking, minutes in in, as we, introduce this. So that covers what we've recently introduced in 8.o, what we're planning on releasing just after eight dot o. And, you know, now I'll, pass this over to Tim to talk about what the new capabilities are gonna be inside of Capella. Great. Thanks, Jeff. And, you know, those those eight dot o features, flow right into Capella as the as the core engine within Capella. So for those, you know, when we launched eight o, it was ready the same day within Capella. For those who are ready to move or building new applications, it's it's already there for you. And most of you probably, if you're not already using Capella, have heard or know about Capella, our database as a service. And we really wanna talk about how, the journey it's been on and and how much it's evolved since the the beginning. It it's come a long way. So, what you can see here is that we're really proud of the Capella adoption. Capella has been in market actually for we're coming close to six years. So in terms of it, being a maybe a new offering or something that, some customers haven't really looked at. It's quite a mature product at this point. In fact, 30% of our customers, are using Capella for their applications. We're over 700 clusters, and that doesn't include the thousands of free tier clusters that that are being deployed all the time. Some very big deployments. One customer using 26 nodes, for their, their application and another customer has a entire footprint over 200 nodes that they've deployed on on Capella. Right? And you can see some of the the kind of marquee names here on on the right, of customers who are trusting us, with our database as a service. Example, Western Union. Early to Capella, they've been using Capella for many years. Surely, you've heard of Western Union. They operate in, you know, over 200 countries. They, are the, you know, money transfer, network that has been around for over a hundred years, has 4,000,000,000, accounts and digital wallets, within their system. And, you know, they began on Capella, started with three applications. They are now up to 11 mission critical applications that they run, including profiles and wallets and things like this, all trusted on Capella. Another example oh, just real quick, is is Aptos. Aptos is, a a retail point of sales, platform technology platform used by Trader Joe's and, companies like Tommy Bahama and and and many others. They're a marquee retailer. They use Capella along with our, mobile side of the house, and have built a point of sale solution that because it's using our mobile product, works even when the Internet goes down. So they're trusting us with their most important clients. And because it's Capella, the onboarding process is very simple. They have a new client that they wanna bring on the platform. It takes a couple seconds. They deploy a new, a a new cluster for them, and they're they're off and running. So very simple. Let's go to the next slide. So for those who aren't as familiar with Capella, I'll give a quick, high level of the entire kind of, you know, nature and value proposition, right, in terms of architecture. Key two main key components. Right? There's the control plane, and the data plane. The data plane is what's isolating the the data, for customers within their accounts, deployed across one of three cloud providers where where they can deploy in we've got 75 regions across the three major clouds. And then the control plane, which is used as a single plane to control all of your clusters independent of where where wherever they're deployed, whatever cloud they're on. You have one screen that's, to to manage your, databases behind the scenes. Right? In terms of operations, all the great things of Couchbase Enterprise, bidirectional XDCR, right, very easy to migrate from from server for customers who wanna do that. A lot of automation built in in terms of maintenance, scaling, turning on off, which is great for, early stage development. You wanna turn it off on the weekends, you just schedule it to be off so you're not incurring dev costs. The automated scaling capabilities, and and our traditional multidimensional scaling really makes it easy for customers to manage how it's deployed. Right? You can do that with a few clicks. And, recently talking to a customer, they were saying that they moved from enterprise edition to Capella, and we're leveraging our Magma engine that Jeff mentioned, our our new high density storage engine. And the combination of those two things allowed them to save a million dollars a year just in reduced, cost and and scalability and, use of the platform. So that's great. For developers, it's interesting. It's very easy for them. The data tools connection, they get a lot of different choices. We've just released, released our data, a REST API. We'll talk about that a little bit later. But just lots of different ways for, developers to engage with the platform and then also control, again, managing the platform. Our management API, which we released, I wanna say, about eight eighteen months ago in the first version, we just continually rolling thunder, continue to enhance the capabilities of the management API. So for customers to control their systems, it makes it very easy to plug it into their tools, and control the the database, the cluster, and all of the different components. Right? Built in monitoring, guardrail safety. We've made a lot of improvement on billing reports, giving customers more build visibility into their clusters, what's going on, what everything's costing. And then from a security standpoint, you know, it it's always been from the beginning, this has to be a secure solution. So BCP peering, doing a lot with private links for the clouds. Right? Multi factor authentication. And then in terms of compliance, it's just continuing to add on different cloud compliances. We've got most of the major standards including including, ISO 20 just 27,001, which is one of the most complex and most difficult to get. We secured that I think it was over a year ago. So very secure, very compliant system there. And at its core, right, as I said, you know, it it's got the core components of Couchbase in it, but really making it easy for customers to set up, to manage, to deploy. And so here, just kind of a quick highlight of, you know, what are those core tasks and responsibilities and how many of them are within Capella been automated in terms of setting up environments, securing the environments, making it easier to to do the disaster recovery failover scaling and all of that. It's it's, you know, it's all in there. So now as we look to the future, what is the goal? What are what are we trying to achieve with Capella? Right? It's already it's 75 regions and it's very capable. It has analytics. It's got mobile. It's got our AI services in there. But there's still room to for improvement. And so how we're thinking about this is a 10 x philosophy. How can we make it 10 x easier, better, faster, and more reliable? Right? That's the goal. And so we'll talk a little bit about some of the things we're doing in that area. So in terms of making it easier, how do we make it easier to scale? So one of the first things that'll be coming is, auto scaling of IOPS and compute. So within your cluster, you can you'll be able to set this up. And so, individually, a machine, it will automatically make it faster if, the workload, demands that. Right? Improving the, conversational experience, the AI, the natural language, the way that a developer can work with, Capella, that is going to be an air another area of of focus. And I mentioned, the REST API. Right? Making it easier for teams to just plug into Capella and get data in and out. So this is easier for developers or you no code applications that you wanna plug in and get data out of Capella. This is going to be the tool without having to deploy SDKs. In terms of making it better, query monitoring, workload monitoring, improvements there to make sure you can you can take, actions, on any changes that might be needed. Jeff talked a lot about point in time recovery, so expanding our capabilities there. The third point, multi and hybrid cloud XDCR. So this is really gonna be great in terms of extending what we already achieve in terms of high availability. This means the customers will be able to XDCR between cloud providers and between, self managed deployments and Capella. Right? So continuing to advance that, make that more robust, giving us more flexibility. And as we saw, was it a month or two ago with, challenges with AWS and other cloud providers going down, you know, the more we can build, resilience and high availability into Capella, the better for those mission critical applications. Improving the the, the RBAC controls, and then in terms of getting faster. So Jeff highlighted the instant data scaling. I got a whole slide on this, so I'll just cover this. We'll we'll get to that in more detail in a second. But faster read and writes. So using m NVMe to make sure that, we can enhance those, disk, cache speeds and low latency. And then faster regional expansion. Being able to add more regions to Capella beyond those kind of 75. I think we're at 80 now, in fact, but just making it easier for us to to do more and be in more places. And then and the reliability areas we're looking at is making it easier to patch, Capella so that way we're we're enhancing it, fixing things, making slight changes behind the scenes, things that that are really easy to do, improvements in monitoring. Right? Making sure that we're staying on top of things. And Jeff talked about the open telemetry. So, you know, customers get more understanding not just of what's going on in the database, but that connection between the application and the database. Right? So those are the kind of the things, in the major areas that we're looking to make Capella 10 x better. And so let's just dig it a little bit more into the the data scaling. Jeff showed a little, chart there. But, essentially, what this means is, with our kind of a a new version of the of the engine, what we'll be able to do is to, you know, go from, let's say, three three nodes to six nodes. And that rebalancing process, which as as we we know today, means that all the data is copy over and it it's a, a process that takes a lot of time, we're gonna be doing that as a file based mechanism, and that's gonna radically accelerate, what it takes to scale nodes, the number of your nodes. And so if you can do that radically faster, you can do a rebalance, let's say, from 10 terabytes going from fifty hours to three minute thirty minute. This is kind of some of the testing that that we're seeing internally. It also means that you can, you know, reduce your recovery times. It means your index is built faster. It means that you can lower TCO because you can run smaller clusters, without having to plan as much for peak loads because we know you'll be able to scale up real quickly. Right? So smaller disaster recovery options too. And this just means overall, efficiency in managing a fleet. And so with this, by being able to when we get get here, what this will mean is that Capella becomes the first, you know, truly elastic, rich database as a service. We know that, you know, DynamoDB has auto scaling, but it's really just a KV engine. It doesn't offer the full breadth and capabilities that Couchbase does. And so this is going to be, we think excellent for customers who are very excited, for this. And this is, in the works. Time lines, you know, this is probably twelve months is is what we're kind of estimating. We'll we'll see. But, this is this is gonna be a big one and, but we're very excited about it. And with that, I'm gonna turn it over to, Mark Gamble who's gonna take us through, our our mobile offer and the latest updates there. Excellent. Great. Thanks, Tim. Appreciate that overview. Well, now, as Tim said, I'm gonna focus on Couchbase mobile, and I'm gonna start by highlighting our decade long journey of innovation and leadership in mobile and edge support. As we look at the, the chart here, you know, since 2013, Couchbase has been, delivering innovations, and we were first to market with really, many groundbreaking capabilities. We introduced the industry's first embedded NoSQL database and sync solution powered by Couchbase Lite and sync gateway. And then since then, have just kept building on that. Our engineers have continued leading the market in edge support, rolling out features like, built in sync conflict resolution, peer to peer synchronization for data sharing without the Internet, c API support that enables Couchbase light to be embedded to any platform, and fully hosted back end, sync services with Capella app services. And, all of this forward thinking and planning is paying off now over a decade later. Our state of mobile adoption remains really strong because of awesome capabilities. More recently launching, on device vector search for offline first edge AI. We were the first to offer, in fact, vector search both on device and in the cloud. And earlier this year, we also launched edge server, and that's for powering apps and resource constrained environments, in Internet dead zones. Really perfectly ideally suited, for restaurants, planes, warehouses, these kinds of things. We'll talk about that, a bit more. And we also, recently announced JavaScript support in Couchbase Lite, which enables offline browser based web apps that can not only synchronize with the cloud, but also with other Couchbase Lite mobile applications, just opening a whole new level of different use case, possibilities. So because of this forward thinking engineering and, constant innovation and edge support, by last count, over 35% of Capella deployments include mobile. That means they leverage Capella app services. That's the hosted mobile sync back end. And because mobile is a strategic focus for us, we plan to continue delivering edge and mobile innovations today and into the future. And to do that, our vision has fundamentally always been to build the most developer friendly cloud to edge data platform with offline first capabilities for, for AI. So these are our three strategic, priorities, which include ensuring customer success. We're heavily focused on product quality as well as, product support and product stability. And our plan is to really double down and lean into enterprise grade features as well as prioritize key customer feedback. What you, the customers, tell us, we're listening. And, that helps us guide product direction. We're also gonna continue investments in offline, the offline first market, right, by taking a two fold a two fold approach. First, addressing platform gaps, especially those that hinder, adoption. And, secondly, improving developer experience with a focus on easier and faster development, for, mobile and edge developers. And then finally, we wanna showcase thought leadership in mobile support and edge AI as we expand with, new offerings like the vector search and, edge server, for example, and, continuing innovations. So as we roll to the next slide. So our roadmap is aligned with our, three strategic priorities, and our first priority is customer success. As I mentioned, you know, we're we're listening to customer asks. It not only strengthens those partnerships, but it also makes the product better. For example, we just announced x DCR support for mobile clusters, and this has been a request from several of our large customers. And this is about specifically enabling active active bidirectional replication or, sync gateway or app services clusters. And this, brings with it zero downtime failover between those clusters for app services or sync gateway. Those major enhancement allows for seamless disaster recovery. But it also brings additional, benefits as well as the ability, for example, to seamlessly and easily migrate mobile deployments or expand them between clusters, and to make it easier and faster to expand out, into new global regions. So this is a big step forward for multi region resilience. Eventing support, another development. This was also just announced, for Caltech's mobile, and this allows a support for Caltech's mobile with eventing. And this allows Caltech's mobile developers to write custom functions for transforming data as it is synchronized to and from the server. And this gives them the flexibility to morph and, decorate and mutate data based on data events. So in the context of AI, what does this mean? Well, if a developer wanted to encode data with vectors from a specific model for semantic search on a mobile device, but that mobile device doesn't have access to the model. Eventing functions can actually call the model in the server tier to vector encode the data when it arrives, and then synchronize it to the mobile device where the vectors can be indexed for on device semantic search. So it's this kind of flexibility we think that developers will really leverage to to, meet, the most stringent requirements for, edge support. And we're also delivering private endpoint support in app services on AWS and other CSPs, planned for later this year. On enterprise grade enhancements, we're investing in high scale application readiness, which includes partition channel indexes and options to disable all docs index to reduce memory overhead in large deployments. It's more efficient memory usage. Yeah. And we're building a distributed concurrent resync capability, which is gonna significantly reduce the time and system load for, large scale resynchronization operations. So they're they're gonna be reduced by an order of magnitude. On, developer experience, we've revamped the onboarding experience for, free to Capella free tier users of app services with a guided tour. So not only now is it free and you can use it in perpetuity with a free Capella database as a service, the guided tour will remove friction for your first time experience. It'll guide you through all of the steps to, get started with mobile, in Capella. And we also shipped major query improvements in Couchbase Lite, like array indexing, support for unnest and partial index support. And this allows, essentially developers to work more easily with complex adjacent data, leveraging more, and deeper, more complex queries with, SQL plus plus. And then on product stability, sync gateway now supports configurable caps on rev cache memory, and it helps safeguard against out of memory issues in production. And we lock persistent logs for app services. So logs can now survive node crashes and and aid with diagnostics. And on edge leadership, we've launched a JavaScript SDK for Couchbase Lite, an exciting new delivery. And this brings full offline capabilities, directly to the browser. It's been a long standing, long anticipated feature, with the support of developers building web apps with JavaScript frameworks. So as we move forward now, we can also talk about, the ability here on, platform expansion. So the, JavaScript, again, allows for that ability to embed the database directly to the, browser based application. And then, again, that can synchronize, not only to the cloud, but with any other, application that has to be running couch based like. On cross platform mobile support, we offer ionic capacitor and react native plugins that are supported by the Couchbase community. But we think these are gonna be very popular with, cross platform developers who can now, embed Couchbase Lite directly to those applications. We also modernized the Couchbase Lite iOS Swift API. We now support, Swift UI. Right? That makes, it much easier for the you Swift developers out there, to, you know, be better aligned with the Swift developer ecosystem, utilize the SwiftUI to build, your iOS applications with Calcrest Lite, utilizing the most modern, development framework, for, these these apps. We're also actively developing quick start, apps for Swift and Kotlin. Right? Again, going along with the quick start app for, app services, this is a notion to help developers just get through those first initial steps, get up and running with offline first sync in just a few mouse clicks and a few minutes. So, our first quick starts for the the SDKs will be for Swift and Kotlin, but we'll continue to add support for other platforms as we go along. And on peer to peer, this is a long standing cloud based mobile feature, a big differentiator for our mobile deployments. Now we're very proud to have rolled out, advances with this feature that now include out of the box device auto discovery. One of the most complex, parts of any tier to peer synchronization, is implementing, that discovery mechanism. How did the devices know that each other are there? We've now made that an out of the box capability, that is auto completely automated. And this includes also mesh support. So as, users come in and out of the peer to peer network, mesh support will adjust, that that, network, to accommodate this. And then coming soon, also very, very, widely anticipated, we'll offer Bluetooth support out of the box. And what goes with that is also network auto switching capabilities, and that's for uninterrupted synchronization. If your Bluetooth connection goes out, auto, the auto network switching or network auto switching will automatically, switch over to the next available network. Users will not even be aware that there was any blip in, in synchronization. So exciting capabilities coming here. Okay. So then what finally, when it comes to, market expansion opportunities, we wanna do so through thought leadership and key differentiation. And this includes, among other things, you know, those recent advancements, that we've talked about, probably most notably, Edge Server, which now adds another dimension and level of Edge support for, mobile and edge application deployments. And, of course, vector search on device, which is a, a key differentiator for us that enables, edge AI. So let's talk for a minute on the next slide about Edge Server. This is the the lightweight database server that's designed to power apps and Internet dead zones on resource constrained hardware. And this would be hardware that's otherwise, you know, too small or underpowered to run a full blown multi node database server cluster. So the small footprint of edge server and minimal resource requirements really make it ideally suited for apps in isolated environments. So this would be like, airliners, restaurants, retail stores, warehouses, use cases where the infrastructure and IT staff are minimal, but there's still a need to power applications, when the Internet goes out. So, Edge Server can run on the smallest of hardware. It offers offers a simple rest API, for web applications. And because it's part of the Couchbase mobile family, it can also natively sync data downstream to Couchbase like clients, as well as upstream to Couchbase server or Capella in the cloud. So a typical use case for edge server would be, an airline seat back ordering system for in flight use when the Internet is unavailable. Edge server runs onboard the plane, allowing data to be processed using the aircraft's minimal computing resources. And cloud to edge data sync handles data consistency. But when there is connectivity, perhaps when the plane is at the airport, data can be synced down from Capella with precision to each plane as connectivity allows, setting them up for the next offline first experience for their passengers in the flight. So a real key, capability here in edge server. And, kind of to close out my portion of today's talk, we wanna reiterate our forward thinking focus in AI alignment. You know, again, we were the first database platform with cloud edge vector search capabilities. Vector search is a key requirement to build GenAI and agentic applications. And with vector searching Couchbase Lite, those apps can run at the edge without the need for Internet. So key benefits, of this notion include offline first operation. Right? You know, semantic searches and applications are are always available irrespective of Internet connectivity and could take advantage of on device data that's happening immediately. You can also reduce data transfer costs because everything is local. You don't have to worry about, transfer costs all the way to the cloud and back. Cloud AI model costs and operational costs by doing query processing at the edge, it can re really reduce overall costs. You're distributing that load. And then on privacy, you can personalize search results with retrieval augmented generation and semantic search or rag and semantic search at the edge without compromising privacy. None of that sensitive data has to even leave the device if you don't want it to. And you can provide faster, searches, arguably the fastest searches against a local dataset with local indexes and and and a local embedded, AI model. No Internet latency to contend with. It's all running locally. And, it's, you know, Couchbase mobile and on device vector search that enables all of this. And when you combine that with Couchbase vector search support in the cloud, developers can build apps that have the flexibility to perform similarity search, reg, and, agentic type applications across web back end applications as well as at the edge and even, you know, choose opportunistically, where to do that processing, you know, based on the complexity and available, resources and, horsepower. So a lot of really amazing things, you know, with these latest capabilities. Caltrans builds on on that over decade of mobile database innovation to affirm our place as really a market leader in mobile sync and edge AI. And it's our ongoing commitment to help customers meet, any speed, scale, and availability requirement that, they deem necessary for their most mission critical applications. And we'll do that by offering the most comprehensive mobile and edge support in the industry, you know, running on that industrial strength, resilient, robust database platform that Jeff and Tim have talked about. So with that, I'll turn it, back over to Tim to talk about analytics. Great. Thanks, Mark. Thanks for the mobile updates. So I'm gonna dig into, Couchbase Analytics, and we're gonna start just kind of with a little bit of a refresher and kind of how we think about analytics and the market, and how we think about the challenges of, doing analytics particularly with JSON data. Right? In a world where, analytics has traditionally been dominated by, relational databases. Right? But that doesn't come without its challenges. Right? If you think about relational data databases and the rigidity of a relation system defined and meant to be there by its heart, can be challenging for, analytics, but also more particularly, JSON data which is inherently flexible and changes, often rapidly over time. Those two that flexibility and the rigidity often are at loggerheads. Analytics systems need to be multi source. Right? You need to be able to pull in not just your own data, but also include data from other systems. So it's also a challenge. There's lots of formats, lots of types of data, so that's something that needs to be overcome. The next item is the slow analytic turnaround. And this is I started my career in in in data warehousing and and and, as a data warehouse consultant. And the, the turnaround has always been there. You take data out of your operational system, move it, export it, transport it, load it into your analytics system, process it, cleanse it, so on. Finally, you do some analysis and then what do you do with it? Right? How do you get it back into the operational system to enhance that application? And and how long does do all of those different steps take? That slow analytic turnaround and no actionability have been around for decades. Right? And so these are some of the things we're thinking about and things we're trying to, overcome with Couchbase Analytics. So in designing our analytics offering, we really try to think, you know, at the core of those challenges, what is the the best way that we can have a robust, quick analytics offering for flexible NoSQL JSON world. Right? So you need to have the the capabilities of NoSQL merge with those, you know, great things about traditional relational OLAP. Right? So it's how do you how do you manage schema availability and scalability and be friendly for developers? At the same time, have the speed, and scalability of columnar storage and separating compute and storage. Right? And massively parallel processing and and and cost based optimization. If you can combine all of those things together, that's what you get with Couchbase Analytics. So it's a real time, analytics for NoSQL, but we can also bring in relational other data as well, of course. So we'll talk about that a sec. Let's go to the next slide. And so here we have kind of a very light architecture diagram. And and so if you look right in the center, what you you can think about is kind of the traditional Couchbase server on the left, the the enterprise edition environment, where you're good doing your operations, you're doing your transactions, you have multi clusters for your high, availability, and all of those kind of things. And on the right, you have your analytics Couch based analytics environment where, as mentioned, you're able to separate your compute and storage, which makes it easy to scale up your compute when you need it, and scale it down. Right? So that helps with cost, but also in terms of, need to do processing because analytic workloads are different than transactional workloads. You wanna keep them separated, where transactional workloads need to be on generally all the time and moving all all the time. Analytics, you can you can run very heavy for a period and then just not over the weekend. There's nothing running. So you wanna scale back that compute. So you need to have those separate environments. And so you need to be able to then move data back and forth in real time, and that's what we can do. So from the the traditional enterprise environment into the analytics environment, data moves over zero ETL gets written in into columnar format for for fast analytics. We can then, do computations, derive new data, do those calculations, and then do things with that data. Again, in real time, write it back into the operational system so you can use that, learned information or calculated derived data in your applications, or you can write it out into other environments to files or your cloud service providers, to to be used elsewhere. Right? So that flexibility. So in terms of ingestion, zero ETL, multi source. You can see here, you can pull in JSON, Avro, Parquet, Delta Lake, Iceberg is coming soon. The kind of formats you're getting out of Snowflake, and and Databricks, if you you wanna bring in that data to incorporate it, you can do that here. Speed and scale, a columnar storage engine. Right? Analytics ready, with compute and storage separation. Real time analysis. So that data comes in, extremely fast in milliseconds, hundreds of thousands of of operations per second moving into the analytics environment. You could do your analysis with SQL plus plus, using Capella IQ or natural language coding assistant. You can write your analytic queries. Right? And then you can do, your analysis with your traditional BI clients like Tableau, Power BI, and and Superset. And we're gonna talk about JDBC a bit more in a second. Right back, I mentioned right back into the, enterprise side of the house or right out to to elsewhere. And then another interesting last kinda point is on Capella, is both with the analytics and the real time operational processing. Capella can act as a very good online and offline feature store. So we have a lot of customers who used this historically for, you know, doing the online, the serving data real fast. Now, with our integrations with Feast, we can act both as the, analytic training side of the house and be the offline. So you can get use both of them in one platform as Capella. And we've also now I've circled and read enterprise because, about a few months back, we launched analytics, on enterprise. So there's a downloadable enterprise analytics version that, customers can run as well. Alright. Next slide. Just a quick customer example. There's a recent customer who, has moved to Couchbase Analytics. And what you can see here on the left is their previous environment. What they would be doing is taking their data out of Capella, doing some custom ETL processes, right, and moving that into AWS. They would then continue to refine and do other steps with it, and convert it into a format that they could bring into Athena, at which point they would then be using their BI tools. Today, they don't need to do any of that. It moves instantaneously into the analytics side of the house and they can can get right into their, BI work, with the same sort of tool. So it's faster, easier, less cost, just a better overall solution for, our customers. Another kind of quick customer example. So this is a customer who, is in the insurance industry and their technology is used to collect satellite data, work with their customers to try and determine their risk and and claims processing. And what they were doing in the past was taking this data, processing it, and they were using MongoDB. And they were just getting to a point where things were slowing down slowing down. We we spoke to them and introduced Couchbase, and Capella and both the as the operational and as the analytics engines to use their solution. You can see here at the bottom right that their, operational query processing went down from four seconds within Mongo to one second. Right? So a huge, huge gain. And then their analytical aggregation processes went down from twenty minutes down to forty four seconds. So, again, a nice, big jump in terms of the amount of speed and scale that they got And all of this in one platform. Right? So very easy to to get all these benefits within, Couchbase, enterprise or Capella. Next slide. And so what does the what does the future hold for our analytics, offering over, let's say, the next twelve months? So platform expansion doing more for us to have this product, in all places. So Azure support, is our next major platform we're looking towards, self managed and within Capella, and then supporting, our Kubernetes, operator offering to make it, for customers who are self managed. Integrations into the, ecosystem to make that easier. I mentioned Iceberg before, but to to get that, support available because we're gonna have a lot of customers have data in that format. The j d b b j d b c driver that I mentioned, for more the customers have a wide variety of BI tools that they want to work. There is a community based driver that is out there. This is something that we will be putting out, to to be a more robust offering for those types of tools. More enhancements to just the core engine. Right? So improving the integration processing, improving how, extending all of the different things you can do with, SQL plus plus, and continuing improvements to the to the engine and the platform. And then finally, in this world of AI that we're in and machine learning, making sure that we're incorporating those capabilities into the analytics side of the house, things like doing vector search, our AI functions, which I think Jeff is gonna be talking about in a little bit, you know, to make sure that it's working on our analytics side, improvements in natural language, and then, of course, connectivity with an MCP server. So that's kinda what the road map on analytics look like. So we're excited about that to see these things come to fruition and, you know, see see what our customers will do with analytics. I'll pass it back to Jeff. Awesome. Thanks, Tim. So as we, start to, round out the entirety of the platform, the latest capabilities that, we've certainly been talking about for quite some time, are now available. So, these AI services are, what we are introducing or we've we've introduced. And the key things that it's trying to address are the following types of challenges. Of course, if you want to, support and and supply better context to your large language model in order to get better answers out of it, right, you need to vectorize your, your proprietary data. And, you know, and and oftentimes, that's a risky endeavor for customers. So, you know, it'd be for a number of reasons. One, it might be in lots of places. So this, access and variety of, of your corporate information, It might be in weird forms like PDFs or images or, you know, even, media different media types. All of that information is, you know, has something vital to offer, in your your AI work. But if you can't get at it and you can't provide it as context to, to your large language model, it's useless. So supporting this particular need for both access and and, you know, and the, and accessibility along the variety of different styles and structures of data is is a key requirement here. You all you know, of course, you wanna make sure that you're supporting conversational style interactivity with not only your betting model, but also your large language model. And then, subsequently, anything that you're using to, to validate, that whether or not you like an answer that a large language model has provided you. So all of those interactions, I think one of the key things to remember just inherently about working with genera, you know, generative AI is that it's text based. And, you know, in supporting all kinds of different, tech structures and being able to assemble and and capture text based data is a requirement of your next generation application or your next generation agentic kind of system. So your interactions themselves will be conversational. You're going to want to capture whatever a large language model says in return, and that's gonna be text two. The challenge is you just can't, you know, predict what it's gonna say. So that's, you know, number three here is how do you end up trusting not only what the model say, but what your agents recommend is their next best action. So you wanna we we wanna absolutely focus on helping, customers reduce, hallucinations and then the negative effect of those solution hallucinations downstream. So if you have you know, one of the solutions here is provide better context, of course. Right? Within the retrieval augmented generation, pipeline or workflow, that context data that is indeed, you know, something that enterprises would provide to to to models, that will help improve the accuracy, of your, you know, of the exchange or of the, response that a model is going to give you. But it doesn't guarantee it. So you're gonna need to make sure that you're double checking your work or double checking its work before you allow it to go on and do something else. Because that cascade could be catastrophic, and we absolutely don't want to, see any customer, stuck in a situation like that. So I think right now, what we're seeing is that's one of the biggest concerns about deploying, a Jetix systems or deploying GenAI based systems is that, they're they're slowly treading forward because of this mistrust of of of AI, and and unpredictability of what AI is going to say. And then, of course, what you're gonna be, you know, the next action that something autonomous might take. You really don't want to, not have any, any any intentionality about what they're what they're supposed to do. Anyway, so, you know, observability and governance, guard railing and validation, those are all key you know, traceability are all key aspects here. I think one of the ways we characterize this is if within your rag pipeline, you do everything perfect and you get to a conversation with the your main line language model, your knowledge base information is great, your vectors are accurate, your conversation is specific, and and your prompt is is specific, you still need to check whether you like what the large language model says and what the agent is going to do next before you allow it to do that. And how will you, you know, double check? Well, you're going to take very likely yesterday's conversation or the last conversation that occurred or the collection of last conversations that occurred and evaluate the current one against those. So you wanna corroborate that the whole activity that you've previously done and liked and accepted, that's the driver for the authorization of the current instance of, of activity that an agent is is taking. So that governance, that, you know, that helps you gain, support here, but you need this traceability. You need the ability to go backwards and see if you like it is number one step. But if you if you don't like the answer, is how can you go back and and and trace back where the origin of the piece of information that might have confused the large language model, where that may have started or originated. And, you know, what was it that you know, what prompt was it in? You know, how did that prompt get poisoned by that piece of data and and then gave you a spurious answer on in return. So traceability and tracing is really key here, and a characteristic that is often overlooked with regard to, building these agentic systems. So our approach right now is to, you know, number one, make life significantly easier for developers, to, you know, give them as much of the data tooling kind of environment that we can so that they can be productive in building agents and tracing agents and whatnot. So we're, you know, working ourselves in, you know, to optimize our experience across our own UI, incorporate natural language capabilities, you know, within all of the the Couchbase product line, and integrate with really popular, the most obvious, AI frameworks and, and and IDEs. So, you know, that again, to make your incorporation of data within your, your agentic work as easy as possible. So this, AI database platform, you know, what we're providing in here is helping you, you know, with that data intelligence, not only the data itself, but intelligence behind it, for your agentic systems. So as we've already discussed, right, some of the things we're knocking down here are, can you get the context correctly? And even if the context is massive and, you know, and you needed to vectorize, you know, terabytes or petabytes worth of data such that you can you know, such that it becomes usable. Right? So supporting high performance at hyper scale for your vector indices, that's, you know, one of the critical elements here. But then also, you know, maintaining both short term and long term memory for those engagements that you have with your models through, through retrieval augmented generation is you wanna make sure that you have a whole memory platform that knows all of that that, situational context while you're conversing with the model, you know, and and that, you know, have that with integrated inferencing, you know, and and engagement with your model. And then we wanna make sure that, you know, the handling of all that data, whether it's the vectorization side or the capture side or the evaluation side, That's still going to be semi structured, unstructured data as we mentioned a minute ago. It's likely to be text. That's great. But that text, we you know, in some cases, it's gonna be very unpredictable. And that's why supporting JSON, in this, your AI database platform. That's critical. Right? Because, you don't know what your models are gonna say until they say it, and you still have to capture it, and you still have to evaluate it to see if you like that response. So finally, ecosystem integrations is making sure that we're a good player across the board. We wanna be the data infrastructure for all of this, certainly. But, you know, we wanna make sure we're integrated with model hubs, you know, orchestration fame frameworks, your rag pipelines, and support, you know, the open, open protocols that are either, you know, have come out like MCP or are under development right now. So when you look at how we've evolved our own platform and our own capabilities, we're getting pretty good at, you know, at at at, adding in AI centered, you know, AI powered kinds of feature sets. So Capella IQ came out in, 2023, and that was our first, jump into offering coding assistance within Couchbase, in in the Capella environment so that you could write SQL plus plus easier just by using natural language prompts. Then we added vector search using the search engine inside of, the Couchbase enterprise, the Couchbase core Couchbase database. And as we did that, we did basically what this you know, the rest of the industry was doing at the time. The same as what Elasticsearch did did, the same as what, Lucene has done, the same as, you know, that you see in a lot of systems. Because the search engine was the most closely, aligned to supporting vectors and and supporting that text based, conversations that you would have with your large language models. So that was phase number one. And then, you know, we after we carried through on that, we, offered, you know, machine you know, we we started work on, supporting machine learning feature store features within, the enterprise analytics and Capella analytics, set of features. We've integrated with feature stores and and feature store, frameworks, even more recently. And then we did some things that really set us apart from the rest of the industry. So we did on device vector search, and Mark's already covered that. But we're the only vendor still that has vector search, you know, storage and search capabilities built for a mobile database. Of course, there's going to be more, but we're the first one to do so. And we see mobile support as being critical for AI development because that's where the users are. I see, you know, my family members using, talking to their telephones all the time, using large language models right now just as search and inquiry mechanisms. But you're gonna build applications that will want that same kind of information as well. And we always have to remember that your mobile phone, your smartphone is where most users engage with technology today. So this is going to be a long standing mobile problem, as we look ahead. We did introduce a a self managed MCP server to be a tool player, in in in the agent in the agentic space. And you'll see as we evolve this, it will be a an important, element to the, AI services as we look ahead. And now the things we started to already cover, the hyperscale vector index and the composite vector index, brand new capabilities that are available across the platform in self managed deployments as well as, inside of Capella to support vectors at scale, because we are anticipating that that will be an issue. Most vector search and vector, vector indexing engines today, the pure plays, other, you know, other implementations are in order to gain their speed advantages, are designed to only operate and store your vectors in memory. And we understand that. We're, we're an original and a phenomenal caching solution, in our own right. So we always have been memory first designed. Okay. So we appreciate that. But as as we know, the way that vectors are going to scale themselves is we need to be players not only in the on the memory side, but also storage on disk where where the the and the, persistence of your vector indices are going to be. And when we do that, you're gonna have to partition all of the, you know, that that those indices, to support and and and, you know, ride alongside where their associated context data is, as well as, you know, being able to scale in a distributed manner. So that's bread and butter for Couchbase. We do that all the time. So and and what we're seeing in the, in the space of other, pure play vector stores is they're either only on disk or they're either only in memory. But our hybrid, vector index or hyperscale vector index allows us to operate both in memory and on disk at the same time. So that's how you get your billion scale economies, and you're working this way. And that's actually when we talked about the performance of our hyperscale vector index against our competitors. Where they were falling short was spilling over the disk and getting vectors that are living on disk and not operating exclusively in memory. So that hyperscale vector index, a big, big, innovation that we've introduced. So and and as we look, you know, today and in the future, is supporting these comprehensive AI services across our platform. So, you know, including natural language query, our AI functions, our agent catalog, which I'll talk about in a second, our model service, our data preprocessing services, vectorization services, and the ability to co locate all of that activity, the data that is feeding and and and, you know, both feeding and and ends up as a result of engaging AI. All of that data being close to the models, alongside the models is going to, produce the, you know, the the lowest latency possible. And so as you'll I'm sure you you recognize, most of the industries is concerned wholly about how fast is that engagement, how fast does the model respond. Well, the model's gonna respond a lot quicker if you shorten or eliminate the network leak that has to that has to take place there. So that's why we're doing things like model on device alongside vector vector search on device or, you know, model execution environment alongside the, you know, the the running of the Couchbase, Capella environment itself, you using our NVIDIA enter enterprise AI capabilities here. All of those make this comprehensive AI services platform possible. And finally, as we look ahead, filling out all of the other requirements that are going to be needed for building, these kinds of systems. It's going to be things like supporting agent memory. So can, you know, can I reinstantiate a conversation, all the good parts of the conversation, with a model again and again and again? And can I double check the agent's work so that I can trust it? Can I trust engagement with an MCP server that I've never engaged with before? That's like a stranger with candy right now. So but we wanna support these, you know, these, future concepts, just as readily. So, you know, support semantic metadata, some more support, you know, sparse vector, capabilities, and, you know, support modern quantization techniques, with our vector search. So there's still a long runway to go, but we feel like right now, there's no other database platform that is capable of supporting all the things that we do in the world of vector search and, supporting and operating your RAG workflows much as as comprehensively as Couchbase offers from the data side. So what we're gonna talk about is the capabilities that we do offer right now, are data preprocessing services, so the beginning stage of your retrieval augmented, generation, workflow, you need to take your your your context data and vectorize it. But in many cases, we said before, that context data is in different structures and different formats. It's, it it says, available as media or available as PDF documents, and therefore, it needs to be turned into text. So that's what this data preprocessing service does is it transforms that information into usable JSON so that it can be fed, you know, both to an embedding model and then also to, your your large language model when you are engaging it with your agent agentic software. So preprocessing service. Second phase of that is, you know and and and second service is the vectorization service. So vectorization service takes your JSON data and the other data and, you know, engages in your your favorite embedding model to build your vectors about your contextual data, your proprietary data. The reason why we're doing that, it's you know, and privately hosting it is so you keep your private data private, but you're vectorizing it so you can grab specific contextual elements from it and supply to a prompt, to a, your larger knowledge model. So vectorization service does that. It works closely with our model service, which is our access point to all the model frameworks we support as well as the NVIDIA, NVIDIA, inference micro microservices, which is what NIM stands for. And the model service has built in capabilities around, caching results and, you know, and and, using guard railing techniques to make sure that you like the, the responses that your your models are giving you. So model service is how we engage our our our different, GenAI models. And then the agent catalog is how we look at and evaluate not only the code of the the particular sets of agents themselves, but also the resulting engagements that they have as they're running. So that's, you know, storing the tools, you know, that that make up the agent, the prompts themselves, the tracing information that, you know, of of all of that. So sort of the lineage and metadata about your agent operations, is is what the agent catalog supports. This operates on, the Capella environment right now. It'll operate eventually on, you know, whatever deployment of Couchbase you're, you have. So it'll take and and and coordinate with the operational, the analytic, the mobile, and the AI and vector data that you want you you wanna have in that single self contained platform. So the AI functions will help you with the agent catalog to make your your, to do that things like semantic analysis. Do you like the response or not? Is it what you expected? Those kind of, activities is what AI functions do. And then our own MCP server for those, you know, agents and tools that you create so that you can make them available to other agents in your environment. Anyway, so this is fundamentally what the, makeup the AI services as we have, today. And so think about the different kinds of use cases we're going you know, we're able to support with all of these capabilities in a single platform. Because we've talked about we have your operational data, we have your analytic data, we have your user based, you know, mobile data or your IoT based mobile data. We have all of your the knowledge, you know, derived and and needed for vectorization and conversation and engagement with your AI models, all in a single unified platform. So we can create things like a hyper personal personalization agent that, you know, we we know, one of our cruise customers, right, is building these kind of systems out so that they can personalize not only the onboarding process, but the booking process, you know, when you're when you're online or the excursion, you know, exercises so that your experience when you're onboard and when you're in ports of call is so phenomenal that you'll wanna do this again and again and again. We've already got customers who are, you know, are are are breaking ground, in this area, straight away. And then there's, you know, things like intent based agents that might, you know, improve customer satisfaction. Right? They figure out, you know, let's say, you know, stock management in, in a retailer and say, okay. Well, you know, the, the Cheerios brand from General Mills cereal is not available. But is there another, you know, oak based, cereal that I can offer to that particular customer in that particular moment, or make sure that, you know, there's some kind of alternative. That kind of, you know, similarity searches is, we're seeing quite a bit of. Supporting, you know, analytics and telemetry information. So we've seen, you know, you know, identify know, KPIs that might be improved or KPIs that situationally, perhaps weather or something like that, are affecting negatively and then reacting to that. We have, you know, one of our, telco providers is doing exactly this is, you know, figuring out where where their, their mobile service equipment is, you know, is either under stress or in distress, like, let's say, it got hit by lightning and figuring out what, you know, how to reroute calls, how to re reestablish connectivity, you know, to all of their, supported, five g lines. So that, you know, and and, you know, report back out on the health of the network and what's happening in the network in real time, you know, real live use case here. And then finally, also, you know, supporting workflow style agents. Right? So, you know, accelerate, common workflows like, you know, ticketing systems or report generation systems or integrating with Slack. Right? Even, you know, capabilities there are are really popular. And I think, overall, what we're seeing, at least right now is customers are finding a great deal of success in using GenAI for coding assistance. Right? Coding and authoring. We see that a lot. We see a lot of applications that are, in in the area of chatbots. Right? So they've taken that knowledge base, they've vectorized it, and then they allow users to engage with a smart chatbot who knows about all this information directly. We did that with our documentation, for example, and, you know, as a second type of early Gen AI application. The third type is translation applications, either voice to text kind of translation, or, you know, different spoken language to different spoken language or written languages. That kind of translation capabilities, GenAI is covering very, very effectively. Now what we're, you know, we're still seeing in its, you know, in its earliest stages are these workflow enabled or, you know, core knowledge based kinds of enabled, at real agentic systems. We know that they're out there, but, you know, they're we also see a lot of debt still hesitancy about, trusting and deploying, these kinds of agents, in production. So, you know, customers that need this kind of flexibility, that need these kind of reassurances, that need trustability in their agentic systems, we think that, the the the the Couchbase AI services across the our, AI database platform is going to be very, very useful for you. So then as we look ahead to what's next is continuing on the path of supporting agent memory. Right? So, you know, this allows us to keep context from session to session. Right? That's an ongoing problem with working with a large language model. We wanna make sure that we're storing both your structured facts and annotations, so that your your the reuse of your context can be much more effective. We wanna make sure that, multiple agents are able to operate and know about common data, common information, you know, in in a in a, you know, a shared space or shared working, working memory. We wanna do, you know, lexical and vector search at the same time, you know, and you, you know, and incorporate specific metadata, you know, all at the same time. We want this to be privacy aware and user aware so that, you know, you know, we support, purging, you know, your your, memory that, you know, have you know, so we can support things like the right to be forgotten, right, which is, you know, a, oftentimes a a legislated or a legal right. And then we wanna support more conversational query capabilities across the board in the platform. So so as to make the building these kind of systems significantly easier, lightening even the load of, you know, having SQL knowledge for, the the usability of Couchbase, but then provide things like semantic views or smarter query generation or, you know, that explainability and guardrailing, say, so that you can explain to someone what that lineage was and where the error came from and prevent it from happening again. Or that you can, you know, build systems that are indeed the guard railing systems of you, the guardrail agents to assure yourself that you trust what the activity of an autonomous piece of software is doing. Those are gonna be the problems, you know, that, need to get addressed in the, in the future, you know, in the world of AI. So I walked through AI. We walked through what the platform is doing. And now I'm going to, pass over to, Matthew Groves to talk about the overall developer experience of what it's going to be like for developers using Couchbase. What are they getting, you know, for for for their activities? Thank you very much, Jeff. And, yes, I'm gonna talk about, our overall strategy with developers and kinda how we evaluate the e ecosystems, the tools, and and what's on the horizon here. So we wanna look at basically three phases of a developer experience or a developer life cycle. So the first one is, is Couchbase going to meet my need? This is the evaluation portion. You know, should I pick Couchbase, and will it do what I need it to do? After that, we go into more of a learn phase. So now the goal here is to kind of go into a deeper understanding of how it works. How do I use Couchbase within my application effectively, you know, beyond the hello world type situation? And then finally, the build phase. So this is how do I get an application that uses Couchbase into production? How do I keep it going? How do I keep it going well? How do I tune it and build production grade apps essentially? So onto our kinda overall strategy. Our goal is to improve the experience for developers who are building critical apps, AI or not, with Couchbase. And so we're thinking about developers who are completely new to Couchbase or developers who have an existing Couchbase app that are evaluating some of our new features, new capabilities, new services. So these priorities are aligned with those three steps, evaluate, learn, and build, of the developer journey. So the main priority of the evaluate stage here is to, establish product fit, help developers answer, can Couchbase solve my problem? And this is done through documentation, case studies, competitive analysis, pricing models, things like that. We're gonna demonstrate value early on to those developers, letting them see features in action through interactive demos, quick start apps, performance benchmarks, things like that to answer those early on questions. And we wanna enable some hands on exploration by providing tools and systems to developers to try the product without any kind of commitment, any kind of red tape or hoops to jump through. So things like free trials, our free tier of Capella, playgrounds, things like that. The learn phase then, the main priority there is to help those developers be proficient with Couchbase. So learning all the various concepts of just basic CRUD, the query, RBAC and security models, you know, giving them some cookbook style recipes, some sample apps, things like this. We wanna allow developers to build hands on product by getting comfortable with Couchbase using all all of our SDKs. We have, 10 plus SDKs, different languages, CLI tools, query workbench, even IDE plugins, IDE integrations, and various AI, frameworks that we'll talk about for sure. And we wanna give them a upgrade path that doesn't cause a lot of friction when they go to explore and learn about features that may not be available through free tier right away. And then finally, the build phase. This is to help developers integrate Couchbase into their real world application with support for SDK frameworks like, Ottoman for Node, FastAPI, Spring Boot, even connectors like Kafka, AI frameworks, reference architectures, things like this. We wanna enable Couchbase to be programmatically managed through, modern software delivery tools for CICD pipelines, things like, Terraform for infrastructure as code. And we wanna provide tools to optimize production workloads. These are things like index optimizer, query optimizer, query monitoring, sizing guidelines, things like that. So the road map here is aligned with the strategic priorities just as it is with evaluate, learn, and and build phases. Some of the recently completed I'm not gonna go through all of these here, but some of the recently completed items that are most important here. One of them is credit card billing. This is allows a developer to have a frictionless upgrade experience. Just put in a credit card number and immediately upgrade right there in the UI, to the DevPro and basic plans. And this gives developers the option to try with the free tier prototype and dev test and then move to production with zero friction. One of one of the, the, completed items that I find very, interesting and important is the dot net entity framework implementation. So this is a very popular ORM, in the Microsoft community, dot net community. And so we've recently released version one of that for Couchbase. And, of course, we, have some future plans for that on the road map, having it available via a b p I o, which is a a web, app platform for dot net developers. We have some connectors for Couchbase within platforms where Couchbase is typically available as a source and destination. So for example, Airbyte or Camel or even Confluent management cloud. We have a managed connector for Kafka, so you can use the Confluent cloud to connect, to manage your your Kafka connector to provide Couchbase as both a source and a sync. Some upcoming items, I think Mark may have mentioned already, but the JavaScript support for Couchbase Lite, this is a very important, release for developers of all kinds. It supports, Couchbase Lite to be embedded in a browser or a web front end application. And, data API for Capella. This is a fully managed service that allows programmatic access to the data in the Capella clusters over HTTP without the need for Couchbase SDKs. And this is a, can address use cases for like, serverless or driverless mode or integrations where the SDK may not be available or may not be appropriate. So that is available there. Couchbase Light React Native. I think, Mark mentioned this in mobile section as well. More importantly, some of the connectors for Couchbase that are coming up is support for Apache NiFi, which is a fantastic data flow automation and ETL tool. It's a it's a like a visual designer that allows you to control the data flows. Couchbase support for that, has been deprecated. We're gonna push out a new update for for current versions of Couchbase. Support for AKA and LightBend, the Alpaca project. This is an open source, streaming, platform implementation that runs on the the AKA distributed commuting distributed computing platform. And there's a lot more on there, of course. So, you know, we're always kind of interested, you know, what are, the tools and and, frameworks that you're building are you most interested in. We're always interested in that feedback. I want to talk about the AI ecosystem. We have a broad range of integrations that are available already, for various phases of the AI or ML pipeline. This is just a snapshot of some of the logos you might be familiar with. You can check these all out right now if you go to our website, couchbase.com/developers/integrations. It shows not just AI, but all the various developer integrations that are available there. We have integrations for popular AI models like Cohere, OpenAI, Gemini, Azure, Bedrock, Cloud, etcetera. And you can use some of these frameworks like Langchain, LAMA index, CrewAI, etcetera, to build your agentic apps. There's also some no code or low code platform supported like, Diffy and Flowize and, n eight m. And again, Couchbase is available as both a data store and a vector store that can be accessed within those frameworks. We launched MCP server, I think it was already mentioned, for Couchbase earlier this year. So you can use, Couchbase within your agents. Also within, Kubernetes and and various other air gapped environments where Couchbase server is running. And we have a road map for MCP where we intend to expose more management functionality via the various suitable tools and functions in MCP. And we're also looking at capabilities to improve the natural language to SQL plus plus queries that we can generate with a semantic catalog type of functionality. And we also have integration within Feast, where Couchbase can be used as a both an online and offline feature store. So lots of exciting things coming up, for developers. And I'm gonna hand it back over to Tim, I think, to finish this out here. Thank you, Matt. That was, great. Good to get all those insights. So, that pretty much wraps up our session for today. So thank you so much for, listening and watching and hopefully, you know, taking in as much of this, great road map conversation that, we can deliver. And if you have questions, reach out to your account teams. I'm sure they'll be able to answer answer questions and get you in touch with the the right people. So thanks again, and, we're gonna call it done. Have a good one.