Table of Contents

• Title and Copyright Information
• Preface
• About Data Science Agent
  • Data Science Agent Concepts
  • Key Highlights of Data Science Agent
  • Data Science Agent: Sample Prompts and Outputs
• Prerequisites to use Data Science Agent
  • Use DBMS_CLOUD_AI to Configure AI Profiles
  • Grant OML_DEVELOPER Role to OML User
  • Add User to the Host ACL
  • Perform Prerequisites for Select AI
    • Grant Privileges for Select AI
    • Examples of Privileges to Run Select AI
• Get Started with Data Science Agent
  • Access Data Science Agent
  • Create AI Profile and Cloud Credential
    • Recommended Models
  • Create an OCI Generative AI Credential and AI Profile
  • Manage Data Science Conversations
    • Create a Data Science Agent Conversation
    • Delete a Data Science Agent Conversation
  • Use Data Science Agent Chat Interface
    • Add Objects to Data Science Agent Conversation
    • Manage AI Profile and Service Level of Data Science Conversation
• Best Practices for using Data Science Agent
  • Recommended Models
  • Associate Database objects to your conversation
  • Ask for clarification
  • Ask in multiple iterations
  • Clarify terminology
  • Follow suggestions provided by Data Science Agent
  • Limit your conversation length and scope
  • Provide context to your conversation
  • Specify a clear objective
• Example of a Conversation with Data Science Agent
• Limitations of Data Science Agent
  • Ad hoc SQL queries cannot be run directly
  • Algorithms supported by Oracle permitted for models
  • Conversation length and scope
  • Error handling
  • Performance and latency related limitations
  • Reuse of existing objects
• Known Issues in Data Science Agent
  • Objects present in user schema are not listed
• FAQs
  • Must tables and table columns have enhanced metadata, such as table names, column names, comments, or annotations to help the LLM understand the data better?
  • Is Data Science Agent available in on premise databases?
  • What are the best practices for table and column annotations?
  • Are there limits on the number of tables or database objects Data Science Agent can consume at one time?
  • Is NL2SQL used behind the Discovery stage?
  • How can we determine which models are best suited for use with Data Science Agent?
  • Can we integrate the Data Science Agent console with tools/agent outside of the database?
  • Will Data Science Agent be available on Autonomous AI Database on Dedicated Infrastructure or AI Autonomous, Data Lakehouse?
  • In a multi-agent architecture, how do you handle permissions and access control between agents?
  • How does Data Science Agent filter object metadata for the LLM when hundreds of tables and thousands of columns could exceed the token budget?
  • Where are resources consumed for AutoML and Data Science Agent? Do they use database ECPUs, or do they run on separate dedicated machines?
  • Can we compare Data Science Agent with Codex using Autonomous AI Database MCP server?
  • Can we detect data drift, analysis, and model re-generation?
  • Can Data Science Agent create notebooks?
  • If the user needs to repeat this process monthly, must they follow the same interactive steps each time, or can the system generate a script that can be run?
  • How does Data Science Agent prevent hallucinations?
  • How does Data Science Agent prevent unauthorized/malicious tool calls?
  • Is any sandbox setup possible with a free trial? Can you guide us on a setup on local machine?
  • Is Data Science Agent immediately available on Autonomous Database Serverless on Azure? Are there any regional limitations, and is it also supported on existing Autonomous Database Serverless instances?
  • What are the best practices to verify the answers provided by Data Science Agent?
  • How does Data Science Agent prevent unauthorized calls, and unsafe inputs or outputs?