The new Oracle Generative AI Agent!

OCI Generative AI Agents is a fully managed service that combines the power of large language models (LLMs) with an intelligent retrieval system to create contextually relevant answers by searching your knowledge base, making your AI applications smart and efficient.

It’s hard to keep track of all the developments that followed the wake of ChatGPT and LLM technologies, but clearly, retrieval-augmented generation (RAG) and AI Agents have proven to be powerful solutions for addressing the complex needs of enterprises

Oracle Cloud Infrastructure (OCI) now offers the OCI Generative AI Agent Service, currently available in the Chicago Region.

Key Features

Supports several data on-boarding methods and interaction channels (chat interface or API)

Creates contextually relevant answers by searching your knowledge base

Provides source attribution for every answer

Offers hybrid search capabilities (lexical and semantic) for more correct answers

Includes content moderation options for input and output to ensure a safe and respectful chat experience

Supports multi-turn conversations, where users can ask follow-up questions and receive answers that consider the context of previous questions and answers

Can interpret data from two-axis charts and reference tables in a PDF, without needing explicit descriptions of the visual elements

All the hyperlinks present in the PDF documents are extracted and displayed as hyperlinks in the chat response.

Documentation link

Get Started: 3 Simple Steps

We can setup the entire framework in 3 simple steps as displayed below.

Create Knowledge Base

An AI agent accesses data by connecting to a knowledge base. This knowledge base stores information in a special way (using vectors) and can pull data from other sources. These data sources help the agent find the right information to use when generating responses. For instance, if there’s an object storage with many data files, the data source connects to it and brings in the files so the agent can use them.

There are 3 options for the data store type

In my case I will be using OCI Object Storage to hold a couple of pdf files with a comprehensive knowledge about the Star Trek Universe.

Create an Agent

Then we need to create an Agent, making reference to the knowledge base we want to use.

This make take some minutes to complete, but once that happens we are ready to test. It’s a pretty straightforward process.

By default there is created an agent endpoint with the agent, but we can create new endpoints. Below are some of the options we have while managing endpoints, including the ability to apply content moderation.

Chat: Test the Agent

In the Chat menu, we simply select the agent and the agent endpoint.

Each response is accompanied by the citations, indicating which parts of the knowledge based were used, allowing us to verify the answer and providing full traceability.