Deploying Your First Generative AI Agent
In this next part, we’ll shift from setup to action and guide you through creating your first generative AI agent inside OCI.
Now Let's Build an Agent
We'll first need to place our documents in a bucket (storage area on OCI). Then we'll be able to build a knowledge base where we will put a link to the storage bucket to include those documents. I don't cover creating the bucket for storage (see the livelabs), but I have put documents in a bucket that include those from the livelabs on using AI; I also added a few of my own. My additional PDFs include the 23ai Release Notes, 19c Administrator's Guide, and my previous blog series on What You Missed at Oracle CloudWorld 2024. The main 
process (see image) when you "actually" build your agent will be the following three steps.
Steps to build an agent (see image):
1. Build a knowledge base
2. Create an agent
3. Test the agent with chat
Creating the Knowledge Base
Creating the knowledge base includes giving it a name, specifying a data source (the bucket with your files in it), choosing which files we want to use in the bucket, and pressing the Create button. Our knowledge base (KB) will start as a brown "KB" circle, which turns green when it's 
ready.
Once we have the knowledge base of documents that we want to use, next we're going to create the agent.
Creating the Agent
Giving it a name, giving it a "Welcome Message" to greet the person using the agent, specifying the knowledge base, creating an endpoint (to use it), and finally agreeing to the LLM license use for the LLM that I'm using (Llama) as a front end to ask questions and getting answers. Creating the agent will go from a brown circle "A" to a green one when it is ready (just like the KB).
Note that your documents in the knowledge base are protected from "prying eyes" as the LLM itself is on OCI and not in the "outside world". Also, note that there are costs associated with OCI. Check with your Oracle salesperson for more information on that.
Next, We'll Actually Use Our Agent (or Test It)
I press "Launch chat" and start by asking it about the release notes for 23ai. I ask: "What are the main bug fixes for 23ai?" This was in a document that I included in the knowledge base. I found getting the answer extremely fast. It finds the correct document, and it also shows citations of the document that are accessed with the exact page number where the information is found. Next, I want to ask some more detailed questions from the documents that I've included. If you read my last blog series, you'll recognize the document referenced.
I now ask, "What happened in Larry's talk at Cloudworld24?" Again, it finds the correct document and quickly answers by letting us know that APEX & Autonomous Database are the tools Oracle will use for all future products. It also talks about the Multi-Cloud Era with Oracle Autonomous Database running on AWS, Google, and Microsoft's clouds. It also shows citations to the document along with the page numbers of the document. I can also click "Show" to show the actual PDF that it used to find this information (see image). Some LLMs can’t cite their sources—but this agent can, showing where it got it and what page it was on!
Next, I decided to try something "fun." I asked it to summarize my blog, but in Pirate Language.
When I read the output, I was amazed. I couldn't believe how well it had put a summary together (I also asked it to answer in Spanish - no problem). Some of the statements in Pirate Language:
"Larry, with his keen eye for the horizon, steered the ship towards the treasure-laden islands o' AI."
"Rich bein' the savvy navigator shared the tales of economic booty gained from plunderin' data with the power o' AI."
An important benefit of using Oracle AI is the Converged Database with Oracle's Security for any database workload, whether it's JSON, the Vector Database, something that includes mapping, or you need AI using algorithms; all of it will work on the Oracle converged database. In my last blog, you saw how easy it is to add a vector data type to tables (vectorizing a blob image - binary large object) so that you can now search the image vector with SQL, JSON, or an Agent. Using Agents with the LLM will generate SQL from a natural language using your preferred language (while writing queries to your database).
Agents can also leverage the 30+ in-database machine learning algorithms to predict the future: including classification, clustering, ranking, regression, feature extraction, attribute importance, anomaly detection, association, and survival analysis algorithms. Oracle can generate, store, and compare different vectors without the data security worries, but can also be used with the most popular LLMs (OpenAI, Cohere, Hugging Face, Llama, Google's Gemini & VertexAI...etc) if desired.
Steve Miranda, Oracle Executive Vice President for Applications Development, showed that there are 100+ Generative AI Features in Fusion Apps and 50-plus AI Agents at CloudWorld. This is the beauty of AI leveraged correctly! AI is being built into every aspect of every Oracle application.
Agents are a FAST way to use the Vector Database. In an earlier blog, I looked at the Oracle Vector database using the Oracle Vector search to look at documents to answer customer questions. See my blog on the "Oracle Vector Database" for more images & and information on this! The question is vectorized, it searches vectorized documents (which were broken into chunks that are vectorized), and then the answer from the document is turned back to English (or any other language needed). You can also include the documents themselves as a result, and you can give locations on a map of where your local stores are all within the same chatbot.
A very good Oracle session showed a great example where they take a picture of a house maybe with the type of roof as one dimension, the decorations of the house as a second dimension, the number of stories as a third dimension, the building material as a fourth dimension and so on (the image shows five dimensions). Then you can search using an APEX App (or SQL - PL/SQL) for a similar house based on the picture of a house submitted and want it to look like (maybe even my current house), but then, I can ask for a certain Price in a certain City. I can search the whole country for a house like mine, or just search a city that I'm moving to for a house like the image that I have submitted. It uses the Vector Distance from the image that I've submitted and finds semantically similar houses based on the vector. The images (houses for sale) are put in the database as binary large objects (blobs - the image itself), but I add a vector data type and vectorize that image (creating vector embeddings for the images stored into the field with the vector data type). Now I can search those images using the Oracle Vector Search.
Leverage algorithms Oracle has, along with GenAI and RAG (to protect your data), to get a Foundation Model for your business. This will be the goal for companies over the next five years. Algorithms are used for business functions needed, carefully weaved with Agents and Chatbots that access vectorized data & images in your database using RAG (to protect your data).
Are you leveraging agents/robots (which work 24x7 - until they break)? Those robots, like the autonomous database that manages smaller databases for you (applying patches or upgrades as you sleep), or robots like Alexa or Cortana. Robots in their many hidden forms are just not going away. Nvidia's Jensen Huang said that Physical AI, or Robots, is what's next, and they are investing heavily in it!
To succeed in the future, you need to leverage the correct type of robots/agents to augment yourself to compete with robots of many types.
Rod Serling's quote on robots/androids from the "Steel" episode, with Lee Marvin, of Twilight Zone:
"No matter what the future brings, man's capacity to rise to the occasion will remain unaltered. His potential for tenacity and optimism continues, as always, to outfight, outpoint, and outlive any and all changes made by his society, for which three cheers and a unanimous decision rendered from the Twilight Zone."
Why do 98% of the Fortune 500 run the Oracle database? Because databases come and go, and Oracle just keeps building on decades of database excellence. See me on LinkedIn, X, or send me an email, but don't forget about all of the great Oracle ACE content that is out there. See the picture at the end of this article with many of those great Oracle ACEs who take the time to share their knowledge with others.
The secret Oracle has for AI, in addition to the Oracle Vector Database, Oracle Agents, Oracle Vision, Oracle Vector Indexes, and Vector Speed, is Security, and it has had the best Database Security on the planet for over 45+ years!
References:
- www.viscosityna.com, www.oracle.com, & technet.oracle.com
- Articles, images, and presentations including: Larry Ellison, Safra Catz, Andrew Mendelsohn, Juan Loaiza, Steve Miranda, MIT, Doug Hood, OpenAI, Augmented Startups, LinkedIn, Google, AWS, Microsoft, ace.oracle.com, Google Gemini, ChatGPT, Database Trends & Applications, Oracle documentation, the Twilight Zone, and Sean Stacey.
- DBAs Next Great Job (presentation), Rich Niemiec
- Larry Ellison, Juan Loaiza, & Andrew Mendelsohn Keynotes at CloudWorld 2024
- Preparing for Oracle 23ai, Rich Niemiec
- Oracle Performance Tuning Tips & Techniques; Richard J. Niemiec; Oracle Press
- On the Opportunities and Risks of Foundation Models, Stanford University (2021/2022); 113 authors
- All companies and product names are trademarks or registered trademarks of the respective owners.
- Rich Niemiec ©2025. This document cannot be reproduced without expressed written consent from Rich Niemiec or Viscosity NA, but may be reproduced or copied for article, presentation, and conference use by the author.
- Neither Viscosity nor the author guarantees this document to be error-free. Please provide comments/questions to richniemiec@gmail.com – rich @viscosityna.com; I am always looking to improve!
Learn from the ACEs of Viscosity & others out there!
Ready to Take Your Oracle Skills to the Next Level?
Go deeper with Rich’s insights by gaining access to past webinar recordings, plus exclusive entry to upcoming members-only sessions through OraPub’s training community.
Check out Viscosity’s event page for upcoming virtual and on-site training opportunities.
SUBMIT YOUR COMMENT