Generative AI can help enterprises accelerate innovation, cut costs, and personalize customer service experiences by automating content creation, streamlining business workflows, and discovering new insights from their data. At the same time, there are broader concerns that generative AI could expose sensitive or regulated data and introduce security risks through the use of AI models hosted by third-party AI providers. 

With Select AI on Oracle Autonomous Database, you can configure and use AI models from a private endpoint hosted on Oracle Cloud Infrastructure (OCI). This approach enables you to maintain data privacy and security by using AI models and VMs that you control in a restricted network environment. Using private Large Language Models (LLMs) and transformers provide additional benefits, such as more reliable response times and predictable costs, among others.  

Benefits of a privately hosted LLM with Select AI

Using a private endpoint has multiple benefits, including:

  • Data privacy and security – With a private endpoint, your data or metadata do not get sent to third-party AI providers. Your sensitive data can be more easily controlled to help you meet enterprise objectives and regulations that restrict how and where data can be processed. 
  • Customization and control – You can host your own fine-tuned AI models to meet domain-specific requirements while models remain in your secure environment, helping you to have greater control over your intellectual property and model changes. Hosting your own models insulates your AI use from unplanned model updates or replacements.
  • Performance and latency – You can achieve faster or more uniform response times while avoiding rate limits or usage throttling imposed by third-party AI providers.
  • Predictable and scalable cost structure – You can leverage in-house GPU infrastructures or Oracle Cloud Universal Credits to avoid usage-based pricing and have more predictable expenses for high-volume usage scenarios. 
  • Integration with agentic workflows and RAG – You can maintain a closed, even air-gapped AI environment (e.g., an environment with a physically and logically isolated network, disconnected from the public internet and other unsecured networks), including AI models, in-database ONNX-format transformer models, databases, and vector stores for RAG and agentic workflows.

Select AI’s private AI model architecture

In the figure below, we show how the private endpoint architecture with Select AI isolates the database and AI models from the internet while enabling secure access through controlled entry points.

You can achieve this secure environment by:

  • Configuring your Autonomous Database Serverless instance behind a private endpoint
  • Running your AI model (e.g., using an OpenAI-compatible framework) on a VM behind the private endpoint
  • Having Select AI communicate with your privately hosted model without internet exposure
  • Keeping all components isolated within your Oracle Virtual Cloud network

Private Ollama model setup architecture

Figure: Private Ollama model setup architecture
 
With this Select AI architecture, all processing occurs within an isolated network environment behind a private endpoint in OCI – giving you a greater control over AI security. Refer to the technical document for additional details and explanations.

How to get started

To get started, refer to this document written for technical database users. It shows how to configure Select AI to work with privately hosted Ollama LLMs and transformers. A similar process applies to other libraries, e.g., vLLM and llama.cpp. Additionally, see Select AI Usage Guidelines for security considerations.

For more information:

•    Setting up a private endpoint for AI models using Open AI-compatible frameworks
•    Video: Getting Started with Oracle Select AI 
•    Try Select AI for free on OCI: Autonomous Database Free Trial
•    Documentation: Use Select AI for Natural Language Interaction with your Database
•    Documentation: DBMS_CLOUD_AI package
•    LiveLab: Chat with Your Data in Autonomous Database Using Generative AI
•    LiveLab: Develop apps using GenAI, Autonomous Database, and React
•    LiveLab: Develop AI RAG Apps with Autonomous Database Select AI