Day One and Beyond - Oracle AI Database 26ai Overview transcript

[music] >> Hello everyone. Welcome to our Oracle AI database series. This session will introduce new features available in Oracle AI database 26 AI. You'll learn how these capabilities can enhance modern enterprise applications by combining advanced AI techniques with Oracle's scalable and reliable platform. All right. So, I would like to go ahead and introduce Killian Lynch, senior database product manager as our main presenter today.

Over to you, Killian. >> Cool. All right. Um Yeah, thanks for the intro. Uh my name is Killian Lynch. I'm a product manager on the Oracle AI database team. Uh I'm PM for both the database and for uh an apps dev toolkit we've got called Oracle backend for Firebase. But, today's session is on Oracle AI database 26 AI. Um I did some looking at I I think I saw the last time there was a Oracle AI database overview, it was version 23.

So, I've got a ton to cover. So, without any further ado, I'll go ahead and start. Um I This is our safe harbor statement. It just says that I might talk about some stuff we haven't released yet, so please don't make any purchasing decisions based on that information. Um so, in about the next 60 minutes or so, we're going to cover what's new with Oracle AI database 26 AI. And then we're going to look at three more key areas being functionality added for the AI bits in the database, for apps development, and for mission-critical functionality.

Uh we'll also take a look at some things to be aware of when it comes to actually upgrading to version 26 AI. Now, I'm sure we're all aware of the impact that AI's already having uh both on our professional life and our personal lives. Um you know, whether it's through using, you know, Perplexity to search the internet or or coding assistance to help build applications or uh being able to get help understanding documents and write documents.

Um I don't think you'd be surprised by the impact that this is likely going to have and is already having on our futures. Um today what we're going to talk about is why the Oracle AI database is best positioned to take advantage of these AI uh functionality, this this AI revolution, if you will. Um the work we've been doing inside of the database is to leverage the advances uh in the world of AI and bring that into the Oracle database world.

Whether that's through LLM integration alongside of the database, whether that's through running vector search queries inside the database, whether that's loading embedding models inside the database, or using uh the private containers that we've got, which I'll be touching on here in just a bit. Um but there's a huge number of features in this space. So big, in fact, that we now refer to the Oracle database as the Oracle AI database.

Now, when it comes to using AI and working with AI in the database, we've got two key principles around building that functionality. That's making AI easier to understand and making AI easier to use. And this is what we refer to as our AI for data strategy. So, throughout the presentation, I'm going to talk about what we're doing Oracle AI database version 26 AI, uh why we're doing it, and how it benefits you. Now, it shouldn't come Well, I I hope it doesn't come as any surprise that the Oracle AI database is architected as a converged database.

I think this slide does a great job of explaining that pictorially, but it brings together a whole variety of rich data types from vector to spatial to JSON to text uh to XML to storing data in in a relational format uh as well as uh the ability to work with different workloads from, you know, blockchain which was introduced in 21c to distributed to AI and ML workloads. Um there are a number of different tools we're going to be taking a look at today um as well as tools like Applications Express APEX for building uh low-code and no-code applications.

We've got Boards, we've got SQL CL uh and then a number of different development interfaces in that space, too. Uh it's available across a number of different platforms, whether that's on premise with things like Exadata and ODA or uh with Oracle Linux uh through various containers, uh through Oracle's Cloud, through Cloud at Customer, and some of our uh cloud partnerships as well like Azure, AWS, and GCP. Now, I'm sure you're also aware that Oracle AI database version 26c AI is our long-term support release.

This release comes with hundreds of new features. Last I checked, I think we were north of 400. Um but this version is all about the AI features and functionality that we've been adding to the database. Uh it's the fact that again right down to the very core we're using these algorithms and we're using these AIs uh to to accelerate and to improve the actual experience that users are having integrating AI and data together.

Now, a couple of things to note about uh 26c AI and I'll talk about this later. But, over the previous quarterly release updates, we've introduced dozens of new features to enhance the database. Now, these include things like improvements to AI Vector Search, new end indexing, improvements to the MCP server, um improvements to JSON relational cloud views, there are SQL improvements. So, there are a ton of new improvements that are coming in each one of these quarterly release updates.

Now, if you've already upgraded to version 23AI, uh all you need to do is apply the quarterly release update that came out in October of 2025, and that will take you to version 26AI. Um if you're on any prior version like 19c or 21c, you would need to upgrade directly to version 26AI. And, we'll talk more about that towards the end of the presentation. Now, I won't read this entire slide to you, um but the Oracle AI database version 26AI is available across all major cloud providers.

It's on Oracle's hardware with things like Exadata and ODA. It's on commodity hardware with Linux x86-64. Uh that MOS note in the bottom right-hand corner is the best place to get the latest information on the roll up the other commodity hardware releases like Windows and AIX. Uh that MOS note is where I get all of my information. It's how I populate these slides. Um so, again, highly recommend checking that MOS note for the latest up-to-date information on the availability of 26AI.

Now, I mentioned we're going to be taking a look at uh these three key areas in AI in the database, apps development, and mission-critical. And I'll start with what we're doing in the space for AI. Now, you're probably going to get sick of me saying this, but yeah, again, this release has been all about AI. Um the biggest challenge that we're seeing inside the industry is how enterprises actually go about adopting AI and integrating AI into existing uh business processes and systems.

Now, historically, the database was built on and is best when working with structured business data, things like tables and rows uh are are where we come from. And we can go through and ask questions like, "Find me revenue by time frame. Find me revenue by product." Uh we can easily ask these business questions, but over time, as app requirements have changed, so is the actual underlying data that needs to get stored inside of the database.

So, we've gone from this table format to a semi-structured format with things like JSON documents to now having, you know, IoT-style devices streaming this this this new generation of data, whether that's video, audio, photograph, social media data. Uh I'll call this unstructured data. And this new unstructured data means the database is uniquely positioned to uh take advantage of these large language models and their capabilities, as well as this massive wealth of information that we've already got stored inside of the database.

So, we're leveraging the LLMs and the data that we've got. Now, this advantage is called AI vector search. And AI vector search is a new way to find and and compare these unstructured data types, these documents, these images, these videos. And at the heart of this functionality is the AI vector inside of the Oracle AI database. Uh vectors are stored as a native data type, just like, you know, JSON is in the database, just just like all of the other data types.

The vector is what actually holds the the meaning of my unstructured objects. So, what do I actually mean by that? And this is done by turning these unstructured objects, these documents, these images, these videos, uh into vectors. And the vector, again, is the representation of what this unstructured object actually looks like. Um so, I'll give a quick example. If I've got three photographs, uh two of them are of, I don't know, one's of a cat, one's of a lion, and then I've got a picture of a piece of fruit.

When I turn those photographs into vectors, they're mapped into what's called a vector space. And then, the more similar those vectors are, and and they're compared mathematically, but the more similar they are, uh the more similar they are mapped in that space. So, then you can go through and allow the database to quickly compare and search through these unstructured objects and and perform lookups to answer questions that you may have about the unstructured uh data.

So, vectors were were um the the vector data type is created through an embedding or an embedding model. These embedding models come from commercial or open source uh providers. Oracle was not in the business of building out embedding models. Um there is no best model. Each model does things better than others. Um as you probably already seen, models are getting released you know, it feels like almost weekly at this point and every other model is better than the last one.

So, there is no best model. Um but you have the ability inside of the database to both load embedding models into the database to create these vectors or we'll also talk about uh a private container where you can post these models um either commercially or or or off the internet uh air gapped in these uh private containers can create these vectors. Um I mentioned the math functions. So, after we take our unstructured objects we run them through an embedding model then we have our actual vectors and and comparing these is done mathematically and and Oracle AI database 26c AI supports a number of the well-proven industry math functions uh to compare the distance between vectors because again, the high level high level explanation is the distance between my vectors it what is what represents how similar or how dissimilar each one of my objects actually is.

Now, my brain thinks in in examples, so I'm going to give you a number of examples throughout today's presentation. Um I you know, imagine we've built an app that allows you know, a a team at a bank maybe with trades operations team uh that wants to be able to go through and find similar past finance trades and exemptions when they are investigating you know, a new client issue. So, the app's goal is effectively to help an analyst reduce how long it takes them to solve a a investigation issue that they've got or find prior resolution patterns, identify the next steps that they should take um so that they can speed up that process.

So, you can imagine we've got all of this information from our past trades and our trade cases. Some of it is stored in structured data inside the database, things like product or amount or region or or or or status. But, we've also got all of this unstructured data. Um things like the issue description that we the summary, the analyst notes, the final outcome. Um and this unstructured data is where vector search becomes especially uh uh valuable.

Uh it lets us search through similar cases not just on um keywords, but on the actual meaning of the data. So, we can imagine we've got all this unstructured data in our incident reports and it holds, you know, information like our analyst notes, the the the the actual description of the incident, uh the outcomes of these incidents, and and a number of other variables. Uh and in this example, what we would take those unstructured objects, embed them and then store them as vectors inside of your database.

So, here in in a in a single SQL statement of five lines, you can go through and perform the actual search of the embedding for a given question and then return the top uh most similar vectors while filtering on existing business data. So, this is obviously just a tiny five-line example, but you've got the ability to go through and build out far more complex AI vector search queries, including building out uh retrieval augmented generation pipelines or or rag pipelines uh uh directly from SQL.

Now, there are a number of uh industry use cases in in in businesses across the world already using AI vector search. I'm not going to go through all of these. You'll also get the slides at the end of today's presentation and inside the slides I've linked all of the blogs uh that talk in a bit more detail about how they're using vector search uh and and then the benefits they've got from using that functionality. Um I'll mention one uh down in the bottom left-hand corner we've got Rappi.

Rappi is like um the Amazon of South America. And what they've done is they've implemented AI vector search directly into I don't know what their actual website is, maybe it's rappi.com, but they've introduced it into their product search. So you can go through um and and search across uh everything they've got inside of their products catalog uh and you'll get back based on uh Oracle AI vector search uh similar queries to your question.

Now the other cool thing I think they did was they used a multilingual embedding model because they're servicing, you know, multiple countries throughout South America and they're speaking a number of different languages. So you can search across any language inside of the search as well and it's all powered by AI vector search on these autonomous databases. Uh so there's a number of different use cases and and industries across uh that are working with uh vector search across the globe.

Um so that is AI vector search. You know, another one of the big features inside of this space is the SQL CL MCP server. Now before I talk about the MCP server, um or the SQL CL specific MCP server, I'll talk a little bit about what MCP is. Um typically I would call this an emerging standard, but with how fast AI is moving I I won't refer to it as that. I believe it came out in around September or October of 2023. Um but MCP is designed to improve how these AI models have these LLMs interact with external data, tools, or or environments that you're working with.

So, the goal is to give, you know, a consistent or a structured way for models to actually access information beyond what they were just trained on. Because the training ends, the model gets released, and then in order to add, you know, specific information, say inside your database, you need a way to access that. Prior to MCP, these tools had to create their own, you know, unique protocols. You had to write the code for that.

Every one of those tools had their own way of doing things, and around that time of of November of 2023, Anthropic introduced MCP as a way to standardize how these tools connect with LLMs. A good example would be what USB-C did for USB. It, you know, consolidates power, data transfer, display. That's how sort of how MCP works, uh standardizing the ability for LLMs to work with a specific tool. So, the Oracle SQL CL MCP server gives AI models a way to work with your AI Oracle AI database data.

So, the MCP server lets an AI do things like list available connections, run SQL queries or SQL scripts, disconnect, and then a number of other specific SQL CL commands. Now, the important part is the model and the AI doesn't actually see your database credentials. Those stay encrypted inside of your local wallets. Now, the interactions by these AIs on the database go through the limits and rules that you are setting and and so nothing's actually happening outside of your control.

You've got full visibility into what the AI is doing. Um you can see the MCP session information uh right inside of the V uh dollar session tables and and all of the actions are logged in that local table. So you're never guessing about what's actually happened. Um the best part about this is because it's a SQL CL feature, I should have explained what SQL CL for those of you who aren't familiar with SQL CL. SQL CL is the modern version of SQL Plus.

So it's the SQL command line tool. Um it's built by the same team that built SQL Plus, by the same team that built by uh SQL Developer uh and and the VS Code extension. But because this is a Oracle SQL CL tool, it means that this actually works for any supported version of the Oracle database. So even if you're not on version 26c AI, you can go through and work with this functionality on version 19c and 21c as well. Um as far as use cases go, uh the use cases are just like AI Vector Search across the board for this.

Um now I've seen some DBAs get great benefits with this helping diagnose say or SQL statements or or performance help tuning, uh indexes, um performance in small database operations. Uh well, I've also seen business users get benefit from this as well by the ability to go through and and analyze uh some of the data that you've got inside of the database. Um so a a very neat new feature inside of this space. Another big new feature inside of the AI space, uh uh we've recently introduced a private AI services container.

Um I mentioned that each quarterly release update we are shipping new features and functionality. Uh this is one of those new features. Um this services container comes preloaded with some embedding models and in ONNX runtime, which means you can deploy this in your own infrastructure. You can deploy this uh on premise and in the cloud, uh completely air-gapped off the internet. Uh wherever you want, but the main point of this is it's primarily used for those air-gapped environments where you can't send data to these commercial companies.

Um like I mentioned, this comes preloaded with a number of embedding models and gives you the ability to go through and then load other embedding models uh that you would like to use as well. Um it provides open AI compatible APIs and then these built-in models can be used to do things like semantic search, like retrieval augmented generation, uh and and and you know, multimodal build-out use cases. The big benefit of this service container is the actual offload functionality that you're going to get from this.

Um creating these embeddings, these these embedding generation is very CPU intensive. So, moving that work outside of the database helps preserve the database CPUs for database operations. Uh and then container handles the actual uh creation across the the the CPUs and the ability to scale out across multiple of these containers. Um like I mentioned, it you use, you know, REST APIs or open AI APIs uh to actually interact and communicate with the container for uh the embedding services.

Uh and for the vector index services, you work with HTTP2 for the calls to implement the actual free vector index SQL statements. But, this is just one option when it comes to working with AI in a private way. Now, another big neat new feature is the private agent factory. This is a low-code environment for business users and and enterprise developers to go through and build to test and and to deploy these I'll call them intelligent agents.

So, the platform connects to the Oracle database, to the enterprise data, to LLMs, first APIs, and a number of other tools, so you can actually go through and build out these governed assisted agents and and workflows. There are a number of pre-built agents that you can work with. Um these pre-built agents are the knowledge agent and the data analyst agent, and then there's a number of templates that you can also use to help get started building other types of agents as well.

Now, this is also a free toolkit of the database built on Ottoman containers. So, just like the embedding container we just looked at, this means you can deploy it on premises, this means you can deploy it in the cloud, wherever you need that to be. Um you can connect to things like, you know, Google Drive, to to your database, to to SharePoint, file systems that you've got, you can, you know, add web pages for crawling or or scraping data, but a very neat tool for building out these governed agents in a no-code environment.

Now, with all that being said, Oracle's approach to AI in the database is very open. So, we're not, like I mentioned, we're not in the business of building LLMs for these embedding models. Our open approach means, you know, we want you to use the best AI models and frameworks that you choose inside of your organizations. Uh whether you prefer more open source or or uh commercially available, that choice is yours. In many ways, this lines up with our our converged strategy and our multi-cloud strategies and our we meet you where you are.

We know that every organization's AI journey is different. Um and the Oracle AI database is built to work with the models, the the the tools, and and the providers uh that your your enterprise deems to use. So, that was four or five features inside of the AI space. Like I mentioned, there's over 400 features inside of Oracle AI database as a whole. And a huge chunk of those is dedicated to the AI functionality. Um I do want to shift gears here a little bit and talk about what we're doing in the apps dev space as well.

Um 26c AI has tons of new features for uh uh AI app developers and and building applications for the enterprise. Um we're going to take a look at the first couple of these in just a moment, being JSON valid views, property graphs, uh and schema annotations. Um we've also added things like JavaScript stored procedures to the database, which means you can use JavaScript directly inside of the database. Uh JavaScript is one of the biggest languages in the world, uh and developers can now go through and execute JavaScript directly inside of database.

Why would you do this? Executing JavaScript inside of database means it's executing uh where the data actually exists, so it improves speed and and the actual security. Uh you can also take advantage of the thousands of of JavaScript packages that already exist. Um so also things like uh data use case domains. Data use case domains are a way to reuse business objects uh or or or create these rules for a specific kind of data.

So it's sort of like a template that says, "Whenever I store this kind of value, you know, these rules or these conditions must apply." So it helps you keep things consistent and reduces code duplication. Cuz you define this domain once uh and then anything that points to that inherits those rules. Another uh neat new feature is the GraphQL support inside of Oracle AI database. So there's a lot of developers out there that work with GraphQL.

A GraphQL is an open source query language or query language. Um it's almost like a give me "What I asked for and only what I asked for" style of language. Um so we now support this inside the database, meaning the database will actually do that translation for you rather than you having to do that yourself, which also means database's optimizer can play a role in this uh to improve the performance of these types of queries.

So again, there's a whole ton of new uh uh developer focused features in this space. Um but I want to talk in a little bit more depth about uh JSON relational duality things. Now, we talked earlier about how the Oracle database is a converged database, or it's a a a multimodal database. So we support all the major data types and workloads inside of the database. Uh but what's different with Oracle database 26 AI is we're actually unifying those data types, not just co-locating them inside of the same database, which means the underlying data can now be viewed and worked with and just and in both JSON, relational, and graph formats.

So, like I mentioned, I I think best in examples, so we've got another example here. So, imagine you and I have been asked to build out an application. And in this case, imagine the application that we're building out allows a a team at a bank to track the different cases that go through the bank through different stages of processing. Now, if I were to ask a developer, you know, how would I build something like this?

They might say, "Oh, use JSON. You can easily model this document, and then, you know, you can just use, you know, REST get and and put operations to to retrieve and update data to the database. You can work with a single document. You make whatever changes you want, and then just write it back to the database." Uh that all makes perfect sense. Um but as the developer starts building this thing out, what they realize is data starts to get duplicated.

So, if I've got, you know, an account that's associated with a bunch of different cases, and each one of those cases uh has a document that repeats that account information, and if I've got the same operations team involved in thousands and thousands or millions of cases, that information is getting copied over and over. But you know what the developer says? "Disk's cheap nowadays. Storage is cheap. Who really cares about that?" The real challenge comes down the road when I need to make a change.

So, what happens if the account information changes? What happens if, you know, the operation team name changes or or some other shared detail actually needs to get updated. We now have to go through and find every case every document that has that duplicated data, make sure it's all updated uh perfectly without missing anything. And that can be a difficult and and slow process. So, the developer thinks, "Okay, what if I store this data as a relational table?

Or what if I make a bunch of tables and the underlying database can handle that consistency for me?" But the trade-off is the developer now loses that simplicity that that that the JSON document gives uh and instead of just working with a JSON document, they have to go through and understand a schema of tables, how everything fits together. Um and the whole point of picking JSON in the first place was for that flexibility, was for that speed of development.

So, in Oracle AI D base 26 AI, this is the exact problem that JSON relational duality view solved. So, developers can now benefit from the speed and flexibility and simplicity that the JSON style development brings while benefiting from the relational storage's uh consistency underneath. So, how does this actually work? You would define uh a JSON relational duality view. A view that represents the document that you actually want to work with.

So, here we define a JSON relational duality view for that banking case built on top of normalized relational tables. Now, when account information gets updated or when I need to change that operations team name, uh the the application doesn't have to manage it across thousands or millions of duplicated case documents. Instead, you make a update in the actual document view. The view tells the database what's changed, and then the database updates the underlying tables properly.

And as a result, every view, every document that depends on that shared data uh automatically gets kept consistent. So, it keep it gives developers the freedom of developing in either JSON or relational, and it gives the application the ability to change the document in whatever way it needs. Then return it to the database, uh and then let the database handle the updates uh for the underlying data without the developer having to know anything about the schema.

So, I mentioned uh this is this works for JSON, but this also works for graph as well. So, what we're really doing here is we're removing the the trade-off. The developer doesn't have to pick between JSON uh simplicity or or relational consistency. With JSON relational to that of views, they get both of those. Now, um as we've already mentioned, one of the big use cases uh of AI is the ability for LLMs to generate and and build applications.

Um AI can generate applications and and answer questions better when it understands the actual semantics of the data that's stored in the database, rather than having to infer those semantics or derive them uh through cryptic column names uh or comments. So, schema annotations is the AI's version of database comments. And and they're necessary to achieve, you know, good semantic results. So, schema annotations are available in both version 26 AI and have been back ported to version 19c.

I believe the RU was 19.28. Um One of the big benefits, I'm going to touch back on the MCP server, uh I believe the MCP server is going to give, you know, developers and DBAs is the ability to actually work with all of these new features that the database is shipping. And one of these features is the schema annotations. You could go through and use the MCP server to actually take a look at all of the the objects you've got in the database and help you build out these schema annotations.

Now, another tool toolkit inside of the uh developer space is Oracle backend for Firebase. So, just like the uh SQL CL MCP server is not a feature specifically of the Oracle database, Oracle backend for Firebase is a feature of Oracle REST Data Services or ORDS, but this delivers uh the ability for native mobile developers, so think iOS developers, Android developers, Flutter developers, uh or web app developers to build out uh uh uh modern web and mobile apps directly on top of the Oracle AI database.

So, there are a number of pre-built SDKs for web and mobile development that give developers the ability to build apps with database authentication, file storage, vector search, uh app trust and and the ability to go through and define uh and who's allowed to read, write, update, and delete the application data, which is called security rules. So, this is bringing a backend as a service style development experience for enterprise developers to the enterprise, so the enterprise means change control of the deployment and and the security of the database.

And enterprise developers get easy-to-use SDKs for both web and mobile development. So, just like the AI space, that was three or four new features in the app dev space. And I apologize I wish a level of uh high-level with missed that I'm going through this. Um there are a ton of things I want to try and get through in in the hour. Um the next area I want to talk about is the mission-critical workloads and and and the features we've added inside this space.

Um so, Oracle AI database 26c AI introduces a whole raft of new functionality uh focused on making the database architecturally more simple and scalable for mission-critical workloads. Um This slide barely scratches the surface when it comes to the features we've got in this space. Um I'll be talking about a couple of these in a bit more detail. Uh another couple I want to just mention on here is raft replication. Uh raft replication we introduced as a way to uh it's a globally distributed database protocol.

Um and it's super handy for bringing high availability to the globally distributed database protocol, which is uh formerly called sharding. Um because one of the challenges prior to uh the raft replication protocol was how you actually handle situations when you've got a sharded architecture uh and maybe for government regulations uh or maybe you want to you know, collocate data closer to users. Um but uh what happens when one of those shards actually goes down?

The new raft functionality is so that you don't have to manage that and script that high availability yourself. Um, you decide the number of replication units that you would like to create and when the database goes through and manage this what's called a replication log that keeps it in consistent with your your units and your leading unit. So, if one of your units goes down, the actual recovery is handled automatically by the database with with minimal impact to your actual applications.

Um, let's see. I'll touch on uh SQL firewall in just a second. I do want to talk about um Oracle true cache. So, a lot of applications out there uh typically rely on some sort of mid-tier cache to improve or to to to speed up response times uh to take some pressure off the actual database uh to improve the experience users have using an application. Um, the reality is most of today's mid-tier caches aren't really a cache in the traditional sense.

Uh they're maintained by application developers. Um, they're not typically consistent by default with the underlying database. You'll for that the developer has to actually go through and and make sure that the cache is is consistent, has hot data um so that the cache is is in sync with the database. Um, what typically happens is the reads that don't need up-to-second accuracy get routed to the cache and then any writes or reads that need the most current information still have to go back to the database because that's that's that's where the real data is where the real source of truth is.

Um, so you get this like like semi-cache tier that's fully managed by you as the developer. So, Oracle True Cache is a lightweight, almost diskless Oracle AI database instance that is deployed as a cache. So, you can point any SQL query at this, and instead of that hitting the Oracle back-end, it's going to hit the cache. So, it's a fully functional, fully transparent data cache. So, if you ask a query of the data that's not already in the True Cache, it's going to automatically pull that from back-end for you, uh from the database.

Uh And when data changes happen, those updates are pushed uh to True Cache in real time. So, the result is you get a cache that stays consistent um without the developer having to manage that functionality themselves. Now, you can deploy True Cache instances uh in in front of the database to obviously improve the availability and and scale out um to handle larger and larger cached data sets. Uh one of the new features that shipped in uh one of the prior quarterly release updates was the ability to actually spool data to disk if everything doesn't fit inside of memory.

You can also also pin specific things to the actual memory as opposed to having that spooled out to cache. Um which means you can you can cache much more data in the mid-tier than you typically could otherwise. So, that's uh uh True Cache. Another one of the big marquee features in Oracle AI database is the SQL Firewall. Now, the Oracle database has had a SQL Firewall for a long time, but it's always been a sidecar engine.

So, this feature actually brings the SQL firewall uh inside of the database. So this is designed to be the first line of defense to stop, you know, internal and external attacks before they actually reach the data inside database. So it's very simple to use. It adds almost no overhead and that's because it's built directly inside of the database. And the other benefit of that is there is no way to actually bypass it.

So it gives, you know, real-time protection against common threats by uh uh blocking unauthorized SQL, whether that's, you know, stopping SQL injection attempts. It's got the ability to factor in session context. So things like client IP addresses or or specific SQL statements when you're deciding what to allow. So the way this works is you would go through and create what's called an allowed list of SQL or or session context like like the IP addresses and then you would enable that allowed list.

So any any statements or requests coming from the application have to pass through this allowed list before they're allowed to actually access the data underneath. So a very new bit of functionality built directly inside of the database. Now I want to shift gears here a little bit. I'm I'm I'm conscious of time here. Uh I want to take a look at our support timeline starting with 11.2, 12.1, and 12.2. Those are currently in upgrade support.

This is a additional cost of option. If you're on one of those versions, we highly recommend upgrading to version 26 AI as soon as you can. Um if you're looking at this saying, "Why does that say 23c AI and not 26c AI at the bottom?" That's just to emphasize again that Oracle AI database version 26c AI replaced Oracle version 23c AI uh back in October of 2025. Now, I'll start with 19c. If you're on 19c, you have premier support extending through the end of 2029.

Uh the extended support goes through the end of 2032. If you're on version 21c, um 21c is our innovation release. Our innovation releases come on the interim between our long-term support releases. So, 26c AI is our current long-term support release. 19c was our prior long-term support release. Uh in in the interim, we released version 21c. The other thing I'll mention quickly is if you're upgrading from version 19c to version 26c AI, you're not just getting all of the new features in version 20c 21c.

Uh you're getting all of them from 21c and 26c AI. So, you're getting close to 600-plus new features inside of the database space when you upgrade from 19 to 26. Uh you don't need to upgrade from 19 to 21 to 26. You would go directly from 19 to 26c AI. Um 21c's premier support ends in the middle of 2027, and the 20 uh the premier support for 26c AI ends in at the end of 2031. Uh the extended support hasn't been announced yet.

I expect to see that relatively soon. Uh that MOS note that I showed at the beginning of the the presentation is the best way to get that information once it's announced. Now, Now, I mentioned one of the big differences uh in this version of the database in version 26c AI is the new features and functionality that we're actually shipping in each one of the quarterly release updates. Uh I would expect you to see you could like see this over the next uh set of quarterly release updates as well.

Um but these are This is primarily driven by how fast the industry is actually moving uh with AI. So, a lot of these new features and functionality are in the AI space, but there's also a lot of improvements to JSON relational duality views. There's been improvements to uh you know, how you actually working with SQL, um new SQL functions, um there there are new SQL features like deep data security was just released in 26.2.

Um Deep data Deep data security answers the question of typically you've got applications that are connecting to the database and you can manage um the the the the user's state of who's allowed to access what inside the application itself. What deep data security does is uh because a lot of AI and AI agents are now accessing data inside the database, deep data security provides you a way to define who is allowed to access the data and what they're allowed to access uh at the actual data level.

Um So again, whole raft of new features and functionality inside each one of the quarterly release updates. Now, a couple other things I'd like to touch on in the upgrades space. Uh we've made some changes to to the naming and the release numbering scheme. Um so I mentioned uh Oracle Database is now Oracle AI Database and with the release of the October version of uh Oracle AI Database 23.26 because of how much functionality we actually shipped in the space for for AI specifically, um the Oracle AI Database went from version 23 to 26.

Now, I mentioned if you're already on version 23, you don't need to do an upgrade to get to version 26, and this is what I mean by that. You are simply applying the October RU. Um, now if you're upgrading to version you can upgrade directly to uh the latest RU. Couple of things to note on the actual numbering scheme. Moving forward, it will be a 23 . year that the uh RU was released and then quarter that the RU was released.

So, starting with the January 2026 uh RU the version number is now 23.year.quarter. So, the April RU that came out is 23.26.2. The July 2026 RU that's coming out very shortly will be 23.26.3. Uh and then following next year, the January RU will be 23.27.1. So, it's a bit easier to understand which RU you're actually on now. Couple of other things to note. Um 26 AI is a uh multi-tenant architecture only. So, if you aren't on the multi-tenant architecture as part of the actual upgrade process, you will need to go through and uh uh go from the non-CDB to a PDB.

And you can do that, and I'll show this in just a slide uh just a second. You can do that during the actual upgrade process. The other thing to note is a traditional auditing is going away. Most people we talked to are already on unified auditing. Unified auditing has been around since since 12c. Traditional auditing was prior to 12c, but I did just want to call that out. All that information is available inside of the actual upgrade guide.

Now, um AutoUpgrade is the supported tool for upgrading the Oracle AI database. It will actually go through and perform the pre-checks for you. It will go through and perform the actual upgrade for you. And it can perform post-upgrade analysis as well. It's also the tool you would use to go from a a non-PDB to a PDB architecture. And it also works for applying the quarterly release updates as well. Now, the Oracle upgrade team, and the team responsible for AutoUpgrade, is run by a great guy named Mike Mike's team.

Mike's team all have blogs where they blog about the best practices when it comes to upgrading. They host free webinars where they'll go through and and, you know, show how to actually upgrade it from non-multitenant to multitenant. They've also got a free hands-on live lab that will set up an Oracle environment for you for completely free. It sets up, you know, a non-multitenant database. It sets up a multitenant database, both version 19c.

And then it gives you a hands-on structured guide and walks you through using AutoUpgrade to upgrade either one of those to version 26 AI. I can't recommend this enough. It's available on Oracle Live Labs. The website is livelabs.oracle.com. Again, I'll share all of these slides with you. So, you'll have that specific link. If you're unfamiliar with Live Labs, Live Labs is a free platform that you can actually go through and test out Oracle tools, not just database tools, but it works with things like their workshops for GoldenGate, there are workshops for Oracle Analytics Server, a whole raft of functionality that you can go through and work with.

One of those is an Oracle AI database 26 AI new features Live Lab. There are over, I think, 20-25 new features. There's a number coming inside this this this workshop. But it gives you, again, a sandbox environment where it sets up the database for you. It gives you a workshop guide that you can actually go through and try out a number of different features. So, if you've got interest in, say, the private agent container, or you you want to try out vector search, I highly recommend trying Oracle Live Labs.

There's also specific industry use case Live Labs. For example, and you can go through and build out retrieval augmented style applications directly using the Oracle Live Labs platform with the the workshop guides. It walks you through the code, it explains the code. Again, all this is is free. It's all set up for you, and I can't recommend it enough. With that, I I'm going to go ahead and stop presenting here. I'll pass it back to uh Student to talk a little bit more about the upcoming sessions. >> Thank you, Killian, for an awesome session.

We really appreciate your time and help, and I hope this was a good learning experience for everyone. Uh now, this slide that you're looking at, if you want to learn more, uh here you will find a comprehensive list of resources that will help you walk through Oracle Cloud, get hands-on experience, and get detailed documentation. Uh once again, you will have access to this deck and links on our Day 1 and Beyond page. That is the page where you went on to register for this webinar.

Uh they will be uploaded there within the next couple days, uh as well as Oracle Cloud Customer Connect. Um What about training for your organization? So, if you enjoyed the session and would like us to conduct personalized training sessions for your organization, do email us at our uh personal uh training email uh down below, and we'll reach out to you. Under the Code Innovate program, uh or it's called Developer Experience now, uh we codeveloped a solution based on your own use case.

Uh and just please remember that personal training is an investment we make in our customers to help them grow and develop their OCI skills at no extra cost to them. Um here is our Day 1 and Beyond sessions and uh upcoming sessions for the month of July. Um you still have the opportunity to attend a couple more sessions this month within our hands-on series, um core series, our database series, our uh security, and what's new in OCI.

Our next database series um will be on Wednesday, July 28th. Uh in this session, you will explore um deployment options for Oracle Exadata, and how to start uh how to start your project with best practices there. I hope you learned something new. Thanks again Killian. Thanks everyone for joining today's session. Hope to see you at our next session. Have a wonderful day.