Build Agentic AI Applications with Oracle AI Database Private Agent Factory - EMEA Tour.pdf

In this session, you'll learn how Oracle's advanced platform allows you to "talk" to your servers, ask natural-language questions about performance, anomalies, and system health, and receive contextual answers in real time. From discovering hidden risks in your database to predicting failures before they occur, you'll see the capabilities Fleet Insights offers DBAs and IT professionals.

1. Build Agentic AI
Applications with Oracle
AI Database Private Agent
Factory
Sandesh Rao
Vice President
Applied AI Technologies
October 13, 2025

2. The following is intended to outline our general product direction. It is intended for
information purposes only, and may not be incorporated into any contract. It is not a
commitment to deliver any material, code, or functionality, and should not be relied upon in
making purchasing decisions. The development, release, timing,and pricing of any features
or functionality described for Oracle’s products may change and remains at the sole
discretion of Oracle Corporation.
Safe harbor statement
Copyright © 2025, Oracle and/or its affiliates

3. 3 Copyright © 2025, Oracle and/or its affiliates
Data is the driver for AI agents,
however providing AI agents with
direct data access, risks leaks,
compliance failures,
and regulatory issues
If you leak health data, financial data, private data you go to jail

4. Why Move AI Near and Into the Database?
4 Copyright © 2025, Oracle and/or its affiliates
Security
Control the security and
auditing of data retrieved by
agents
Simplified Development
Leverage the power of the
Converged AI Database
Architecture to access
any data required by the
agent in any format
Performance
Optimize multiple data
retrievals and Oracle specific
capabilities required by
agents
Available Everywhere
Bring AI to the data and
provide always-on
availability and resilience
Agentic AI solutions with enterprise-grade security, performance, and governance

5. 5 Copyright © 2025, Oracle and/or its affiliates
The biggest challenge enterprises
face when adopting AI is integrating
AI into existing business processes
in a way that delivers real, scalable
value
A no-code agentic workflow builder slashes time to solution from
months, to days or hours

6. 6 Copyright © 2025, Oracle and/or its affiliates
There is a fragmented, rapidly
changing eco-system of existing
agentic frameworks
n8nMS Agent
Framework
Building on Open Agent Specification allows us to meet customers
where they are today without exclusions

7. Oracle AI Database Private Agent Factory
What is Oracle AI Database Private Agent Factory?
Agent Factory is a no-code platform designed to help
enterprises rapidly deploy intelligent agents by leveraging:
•Pre-built Agents
•Custom-built Agent
(drag-and-drop - Agent Builder)
•Free to use with the Oracle Database
Objective
•Empower business users and engineers, to launch smart
assistants, no coding required
•Usage included with the Database license
Features
Pre-built Agents
Instantly deploy agents for
common use-cases
Visual Agent Builder
Drag-and-drop builder
Component Variety
LLMs, prompts, data-sources,
vector store, agents, tools
Oracle Components
SQL queries, Vector Search,
OML, Statistical & Analytical
functions, etc
Live Testing
Built-in console for response
simulation
Serialization
Agent / Workflow saved as yaml
files
Interoperability
Same representation as in-
Database Agents
Security
SSO and built-in Role
Management
7 Copyright © 2025, Oracle and/or its affiliates

8. Agent Factory
High Level Architecture
Pre-requisites
1.Oracle Linux VM (OL8) or a Mac
2.Oracle 23ai Database if you want vectors ,
Knowledge Agent
3.Podman / Docker
4.LLM Endpoint
Agent Factory Container
5.Nginx, Flask, Gunicorn
6.Agent Execution Engine
7.Ingestion Service
Enterprise’s Data Sources
8.Oracle Database (19c onwards)
9.REST APIs (OpenAPI compatible)
10.Web Sources
11.File System
12.OCI Object Storage
8 Copyright © 2025, Oracle and/or its affiliates

9. AI Agents
Copyright © 2025, Oracle and/or its affiliates 9
Agent : An entity that makes decisions , it receives
input, decides what tool(s) to call, and generates
output
Tools : External functions or APIs the agent can invoke
to perform tasks
Memory : A structured way for the agent to store and
recall past interactions, decisions, or observations
Reasoning Engine : Guides the agent in breaking down
goals into sub-tasks and deciding the next best action
Action Executor : Executes the decided tools or actions
based on reasoning

10. 10 Copyright © 2025, Oracle and/or its affiliates
Agent vs Workflow | Decision Making
Agent : An entity that makes decisions - receives
input, decides what tool(s) to call, and generates
output
Workflow : A predefined sequence of steps (nodes)
that executes in a flow, often without autonomous
decision-making at every step
Example : Conversational AI, decision-making bots Example : Order processing flows

11. Getting Started with Agent Factory
Landing page for Agent Factory
Guides user to:
1.Use existing Pre-built Agents
•Knowledge Agents
•Data Analysis Agent
2.Create Agent/Workflow form Scratch
•Using Agent Builder
3.Use Template Gallery
•Has pre-built agent / workflow for
editing
11 Copyright © 2025, Oracle and/or its affiliates

12. Single Sign On

13. Copyright © 2025, Oracle and/or its affiliates

14. Knowledge Agent
Pre-built
What is Knowledge Agent?
Agent that augments 26ai AI Vector Search &
LLM capabilities with Enterprise Data, enabling
accurate, context-rich responses by retrieving
relevant information from knowledge bases,
documents, and web sources. (“RAG-in-a-Box”)
Core Capabilities
•Contextual Retrieval from unstructured
data sources
•Grounded Responses traceable to
enterprise-approved sources with
guardrails
•Integration Ready - connects with corporate
sites, file systems, etc.
14 Copyright © 2025, Oracle and/or its affiliates

15. 15 Copyright © 2025, Oracle and/or its affiliates
Knowledge Agent powered by 26ai AI Vector Search
User Query
Top K matches
AI Vector SearchML Embedding Model
Query Vector
compare
…
Data Corpus
(encoded with the
same embedding model)

16. A user’s natural language
question is encoded as
a vector and sent to
AI Vector Search
1
16 Copyright © 2025, Oracle and/or its affiliates
RAG with 26ai AI Vector Search works like this
16
Why does
my laptop
keep
rebooting?
334216215
AI Vector Search
AI Vector Search finds private
content such as incidents that
match the user's question
Can search for relevant product type
and customer location
2
Incident
Reports
Product and
customer data
User
The user's question is augmented
with relevant content3
LLM
LLM uses the question plus the
content and general knowledge to
provide a better informed answer4
The issue is with
the firmware controlling
the fan. Apply OS update
42 while plugged in, and in
a cool air- conditioned
environment to prevent
overheating

17. Knowledge Agent – Data Sources
•Fully automated data ingestion pipeline
•Crawl
•Parse
•Chunk
•Embed
•Ingest documents stored in:
•Websites
•File system (file upload)
•SharePoint
•Confluence
•OCI Object Storage
•Parse documents in various data formats
•HTML files
•PDF files
•Microsoft Word documents
•Text files
•JSON files
17 Copyright © 2025, Oracle and/or its affiliates

18. Copyright © 2025, Oracle and/or its affiliates

19. Copyright © 2025, Oracle and/or its affiliates 19

22. 22 Copyright © 2025, Oracle and/or its affiliates
Knowledge Agent pipeline
Data Sources Document
Loaders
Document Transformation
(e.g., Text Splitting,
Summarization)
Embedding
Models
Vector
Database
Similarity
Search
LLMs User
RAG
22

23. Structures – Data Analysis Agent
Pre-built
What is Data Analysis Agent?
Agent designed to interact with your enterprise databases,
understand schema structure, and extract various Semantic
Insights automatically using LLM capabilities without
sending the actual data to the LLM
Core Capabilities
•Semantic and data-relevant question generation via
Variation Analysis
•LLM powered explanations
•Automatic Visualization Generation
•Integration Ready - connects with database 19c and
above
23 Copyright © 2025, Oracle and/or its affiliates
Movie
Dataset

24. 24 Copyright © 2025, Oracle and/or its affiliates

25. Agent Builder
Key features
•Drag-and-drop interface – Visually construct workflows by connecting nodes that represent specific actions, tools, or data
sources
•Integration of AI and automation - Supports components like LLMs, chat agents, and data processing tools for building
intelligent workflows
•Custom agent creation - Agents that autonomously perform tasks, make decisions, or interact with systems based on
workflows
•Connectivity - Facilitates integration with enterprise systems, third-party services, cloud APIs, and databases
•Reusable templates - Offers workflow templates that can be adapted to different business needs
•Built in Guardrails – Using prompt guard we add a default set of guardrails to the Agents created
•Jinja Templates in prompts : Easy to use Jinja template format to create connectors when building nodes
25 Copyright © 2025, Oracle and/or its affiliates

26. Copyright © 2025, Oracle and/or its affiliates

27. Nodes in an Agent Builder
Agent Builder
•Language Model
•vLLM - A node leveraging the vLLM (Virtualized Large Language Model) to serve fast, efficient LLM inference
•Agents
•Agent Node - An agent node that utilizes a vLLM instance or OCI services to carry out instructions
•MCP Server - The MCP node is a specialized component designed to interface with AI models using MCP protocol
•Inputs
•Chat Input - A node specifically designed for conversational user input
•Prompt - A node for modifying the textual instructions sent to a language model, customizing the context
•Outputs
•Chat Output - A node rendering or returning model or agent responses
•Data
•Read CSV - A node for importing and parsing comma-separated value files,
•File Upload - A node allowing users to upload files for processing or analysis within a workflow
•SQL Query - A node for executing SQL statements against databases, returning query results to the workflow
27 Copyright © 2025, Oracle and/or its affiliates

28. Nodes in an Agent Builder
Agent Builder
•Data
•Data Analysis Node - A node for importing and parsing outputs of the Data Analysis Agent
•Knowledge Agent - A node allowing users to Ingest content output of the Knowledge Agent
•Vector Node - A node for executing similarity search against databases, returning query results to the workflow
•Basic RAG – A node which allows to use LLM interactions with a Vector node to perform simple one shot RAG functions to
enable as part of the agent flow
28 Copyright © 2025, Oracle and/or its affiliates

29. Agent Builder Execution Flow
29 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted
The agent is executed using
the underlying framework
with the tools split across the
orchestrator agents and
subagents
A dynamic agent or workflow
is created by Agent Execution
Framework based on the
transformed data

30. 30 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted
Multi Agent Architecture
Clear Prompt Instructions
•Define each tool’s capability explicitly in both orchestrator and agent prompts.
Response Format Guidance
•Standardize responses by instructing both orchestrator and agents on the expected format.
Tool Splitting by Semantic Similarity
•Break large toolsets into sub-agents based on topic similarity.
•Example: Grouped all database/storage tools into one sub-agent under an “insights agent” to improve focus and accuracy.

31. 31 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted
Multi Agent Architecture – What to be aware of
Multiple Levels of Orchestrators
•Orchestrators become less effective with too many agents — classic "needle in a haystack" LLM issue
•Use layered orchestrators which support multi-level orchestration
Too Many Tools per Agent
•Agent accuracy declines when it handles more than ~15 tools.
-Complexity and tool similarity worsen performance but the agent framework splits this into multiple subagents
Non-Definitive Tool Responses
•If a tool returns "Response not found," the agent may
-Try irrelevant tools
-Hallucinate answers instead of admitting uncertainty
•Prompts internally make sure to have realistic answers instead of being creative
Granular Agent Decomposition per Query
-Use parallel query execution
-Can be limited by token limit issues with complex queries or verbose agent outputs

32. Model Management
Settings
Generative Models
•OCI Gen AI
•Meta Llama
•Cohere Command R+
•Open AI
•Grok
•Ollama, vLLM
•Customer decides the
model
•OpenAI
•gpt-4o
•gpt-4o-mini
•gpt-5
32 Copyright © 2025, Oracle and/or its affiliates

33. Datasets
Utilities
•Agent Factory to allow import of few datasets for customers use, so they can try out agents, out-of-the-box.
•Use for Data Analysis Agent (Structured Dataset) and Knowledge Agent (Un-Structured Dataset)
33 Copyright © 2025, Oracle and/or its affiliates

34. Prompt Lab
Utilities
What is Prompt Lab?
•Prompt Lab allows users to experiment with different prompts and fine-tune their interaction with various AI models.
•Users can adjust model settings, test multiple prompts, and observe the resulting outputs in real-time.
•Prompts can be saved for repeated use, streamlining future experimentation.
•Acts as a offline prompt repository, offering pre-built prompts to accelerate development and testing
Core Capabilities
•Facilitates Rapid Experimentation: Quickly try out ideas to determine the best prompt formulations for your app
•Model Comparison: Evaluate outputs from different AI models for the same prompt
•Knowledge Retention: Save and manage effective prompts for consistent results and future reference
•Boosts Productivity: Access a library of tested, pre-built prompts to save time and standardize best practices
•Supports Iterative Improvement: Fine-tune prompts based on model performance
34 Copyright © 2025, Oracle and/or its affiliates

35. 35 Copyright © 2025, Oracle and/or its affiliates

36. Copyright © 2025, Oracle and/or its affiliates

37. OpenAPI Compatible REST APIs
Data Source
What are OpenAPI Compatible REST APIs?
•OpenAPI Compatible REST APIs are web services documented using the OpenAPI Specification which is an industry-standard format
(JSON/YAML, version 2.0 or 3.0) for describing RESTful APIs
•OpenAPI specs define endpoints, methods, input/output parameters, authentication and other details, making APIs easily discoverable
and consumable
•Using OpenAPI enables automation, testing, and integration across platforms and tools
Importance for Agent Building Platforms
•Easy Integration: Accepting OpenAPI specs enables the platform to automatically understand and connect to third-party REST APIs, even when
direct MCP integration is unavailable
•Tool Conversion: The platform can turn OpenAPI-defined APIs into agent tools, allowing agents to leverage external services efficiently
•Review & Editing: After uploading the specs (OpenAPI 2.0/3.0 in JSON), users can review, try out APIs, and update descriptions or parameter
details to optimize agent behaviors
•Accelerated Development: Reduces manual configuration, speeds up workflow, and minimizes errors in connecting to external APIs
37 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

38. OpenAPI Compatible REST APIs
Data Source
Steps to add new OpenAPI Data Source
1.Upload OpenAPI 2.0 or 3.0 compatible json specification
38 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

39. OpenAPI Compatible REST APIs
Data Source
2.Once loaded you will be able to review and edit.
39 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

40. OpenAPI Compatible REST APIs
Data Source
3.Try out the APIs and update their descriptions and parameter explanation for tool conversion.
40 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

41. 41 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

42. 42 Copyright © 2025, Oracle and/or its affiliates

43. AgentURL and inline execution in code
Sequence of steps to create an Agent
•Use Agent Builder to build the flow
•Run the agent in the playground until perfected and the flows work
•Expose the Agent using the AgentURL
•Create Authn/Authz for accessing AgentURL
•Take around 5 lines of code and import into python code including the “Agent Factory” module
•This is part of the python SDK which will ship with the Agent Factory
•Start conversation with the agent using the URL and Authenticate the client
•After the initial conversation use the roomID to continue context and memory for future conversations
•The entire execution is treating the Database Agent Factory as a headless node and the Agent could be tweaked
as needed to fulfill future requirements
43 Copyright © 2024, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

44. 44 Copyright © 2025, Oracle and/or its affiliates

45. 45 Copyright © 2025, Oracle and/or its affiliates

46. 46 Copyright © 2025, Oracle and/or its affiliates

47. Evaluation Is Critical for Customers to Deploy and Optimize AI in Production
47 Copyright © 2025, Oracle and/or its affiliates
•Admins require evaluating RAG and AI agents on their own enterprise data, queries and machines at
large scale before launching in production
•Large enterprise investing on AI often require maintaining competitive advantages
•This means that enterprise admins constantly evaluate fast evolving and complex AI stacks
•The high cost of AI investment also pushes for continuous tangible business metrics lifts, and thus
continuous evaluations of new AI innovations internally and externally
•Democratizing generative AI : everyone can build, validate and deploy in production easily

48. Building Evaluations Is Labor Intensive | Step 1. Defining Metrics
48 Copyright © 2025, Oracle and/or its affiliates
•Multiple metrics are required to understand in-production RAG or AI agent systems
•Goodness of RAG and AI agents are multi-facets and sometimes conflicting e.g. accuracy v.s. latency
•RAGs and AI agents require evaluating concepts which are hard to measure directly with a single metric
•e.g., RAG generated answer quality can be evaluated by relevance , correctness , faithfulness , completeness
•RAGs and AI agents are complex software which requires fine granularity metrics for actionable insights
•e.g., retrieval recall and reranker precision in a RAG system; trajectory efficiency and tool selection accuracy in AI
agents
•Additionally, customers also need business/use case specific metrics and guardrails
•Without out-of-box support, users need to write code to achieve all of the above

49. Building Evaluations Is Labor Intensive | Step 2. Curating Evaluation Data
49 Copyright © 2025, Oracle and/or its affiliates
•Representative “cold start” data : Generating diverse, production-aligned evaluation data for "cold start"
systems is hard
•“staging” data not always available due to infra complexity/cost
•user queries are not available
•simulating user queries context is hard. e.g., how user interact with a Q&A system
•Non-intrusive production data capturing : Capturing evaluation data from live production systems required
tradeoffs in performance, security/privacy and data utility
•Multi-layered golden data : Creating multi-level human-labeled ground truth or golden data for complex AI
systems, including intermediate steps, is highly labor-intensive

50. Building Evaluations Is Labor Intensive | Step 3. Evaluation
50 Copyright © 2025, Oracle and/or its affiliates
•Human evaluations are needed for subjective metrics such as Relevance, Faithfullness
•Human evaluations are often expensive because
•# of data in evaluations is in size of 100s to 1000s+
•it needs to repeat many times for any tunning in RAG or agent solution stacks
•evaluations are non trivia, requiring following complex guidelines or using tools to research
•evaluations sometimes require domain expertise, and qualified evaluators are scarce resources
•Human evaluators are often one of most costly teams in all big AI

51. 51 Copyright © 2025, Oracle and/or its affiliates
Building Evaluations Is Labor Intensive | Step 4. Taking Actions
Even with validation data and metrics measured, acting on evaluations is still labor intensive
•Conflicting objectives: multiple, often conflicting, evaluation metrics obscure critical tradeoffs
•E.g. cost and quality tradeoff of different LLMs
•Vast Search Space: manually optimizing in vast parameter space is expensive and sometimes impractical
•Interdependent components: Complex component interdependencies complicate effective action prioritization
data sources
query retriever reranker
- query rewriting
- query expansion
- search strategies
- distant functions
- K in top K
vector
database
- vector index types
- vector memory size
embedding
generator
- vector embedding models
- embedding data types/sizes
- chunking strategy
- chunking granularity
- reranking models
language
generation
Response
- model choices
- prompt engineering
- finetune LLMs

52. 52 Copyright © 2025, Oracle and/or its affiliates
Evaluation embedded into the Database Agent Factory
•A solution that automates high-quality, large-scale evaluation of agent and RAG applications on customers’ data
•An Oracle native evaluation solution, deeply integrated with Oracle databases to streamline productionizing on these
AI products through metrics driven and actionable insights.
•Centralize evaluation data and analysis in Oracle, together with customers’ vectors and structured data
•Predefine both general purpose evaluation metrics and data-centric agent metrics (e.g. semantic correctness of data
agent output, efficiency of a data operation, robustness to schema complexity)
•Generate application specific synthetic data when lacking production evaluation data
•Enable large scale high quality evaluation with minimal human efforts by combining LLM-as-a-judge and human
judges
•Provide analysis, diagnosis and action recommendations to optimize based on evaluation results

53. Agent Specin Database Agent Factory
Leveraging Agent Spec
Agent Spec agents support import/export:
•Users can export their agents to Agent Spec from a variety of frameworks:
•LangChain / LangGraph, Autogen, CrewAI …
•Users can import such Agent Spec files into the Private AI Agent Factory
•Once imported, Agent Spec definitions are automatically converted into runtime-ready agents that can be
served and scaled on the platform
•Ensures consistent behavior across frameworks, reducing integration complexity.
•Promotes reusability and collaboration, allowing teams to share and standardize agents seamlessly.
53 Copyright © 2025, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

54. Agent Spec & Private AI Agent Factory
React Agent with MCP tools
54 Copyright © 2025, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted
The agent is written in langgraph, and converted to Agent Spec
LangGraph Code
Agent Spec
LLM
MCP
server
React
Agent
LLM
MCP
server
React
Agent

55. Agent Spec & Private AI Agent Factory
Code Review Flow
55 Copyright © 2025, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted
The agent is written in AutoGen, and converted to Agent Spec
Agent Spec
LLM
Code
Generator
Agent
Flow
AutoGen Code
Code
Reviewer
Agent
LLM
Code
Generator
Agent
Flow
Code
Reviewer
Agent

56. 56 Copyright © 2025, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

57. Agent Builder Examples

58. 58 Copyright © 2025, Oracle and/or its affiliates

59. 59 Copyright © 2025, Oracle and/or its affiliates

60. 60 Copyright © 2025, Oracle and/or its affiliates

61. 61 Copyright © 2025, Oracle and/or its affiliates

62. 62 Copyright © 2025, Oracle and/or its affiliates

63. 63 Copyright © 2025, Oracle and/or its affiliates

64. 64 Copyright © 2025, Oracle and/or its affiliates

65. 65 Copyright © 2025, Oracle and/or its affiliates

66. 66 Copyright © 2025, Oracle and/or its affiliates

67. 67 Copyright © 2025, Oracle and/or its affiliates

68. Delivery Mechanism
1.MOS – My Oracle Support
MOS Page and Oracle.com page for Agent Factory
2.Oracle Container Registry + Github
Here the Agent Factory image is present on container registry.
Supporting utility files are present on github.
The utility file will help setup:
23ai free container is pulled from container registry (Not Shipped)
Ollama container – we are only providing utility for easy container
setup (Not shipping ollama)
3.Oracle Marketplace
docker pull container-registry.oracle.com/applied-ai/a-
agent-factory:latest
68 Copyright © 2025, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted

69. 69 Copyright © 2025, Oracle and/or its affiliates
Oracle AI Database Private Agent Factory| Quick Start Architecture
Configure
Question / Answer
Users
Oracle Linux 8 or Mac
Oracle AI Database Private
Agent Factory
Ollama
Open Source LLM Service
Oracle Database 23ai Free
Generate
Store Data /
Search

70. Production ready air-gapped architecture
Copyright © 2025, Oracle and/or its affiliates
Store Data / Search
Generate
Configure
Question /
Answer
Users
User
Oracle Engineered System
Oracle Database 23ai
GPU Machine
Ollama
Open Source LLM Service
Admins
Administrator
Oracle Linux 8 Machine
Oracle AI Database
Private Agent Factory
Existing SSO
Identity Provider
Authenticate
Identification /
Authorization
Question / Answer
70

71. Agent Factory - Summary
•Custom Agent Builder:
•Build, Test and Deploy custom agents / workflows for customer’s own usecases
•Components
-LLM Support: vLLM, Ollama, OCI GenAI, Private AI Services Container, OpenAI, Grok
-Agent Support: LLM + Tools + Conversational Memory
-Tools: MCP Server (SSE and HTTP Streaming), OpenAPI Compatible REST APIs as Tools
-Inputs: Chat Input, Text Input, Prompt Node
-Outputs: Chat Output, Email
-Data: File Upload , Read CSV, SQL Query
•Extendible: We provide a guide to create new nodes in Agent Builder
•UI/UX
•Oracle Jet based modern UI
•Chat Interfaces with AG-UI support with support for user feedback, related questions, charts & visualizations, markdown
support and chat history
•Prompt Lab: allows users to experiment with different prompts and fine-tune their interaction with various AI models.
•Interoperability (Open Agent Specification ) – Import agents from external frameworks like LangGraph, Autogen
•Security: SSO (OAuth2), Users & Roles (Admin, Editor, Chat-Only users)
71 Copyright © 2025, Oracle and/or its affiliates

72. Agent Factory LiveLabs steps when available
72 Copyright © 2025, Oracle and/or its affiliates

73. Your feedback is important.
Scan this QR Code or use the
Mobile App to share your
thoughts on this session.
Copyright © 2025, Oracle and/or its affiliates | Confidential: Internal/Restricted/Highly Restricted73

74. Copyright © 2025, Oracle and/or its affiliates 74

75. Questions ?
Copyright © 2025, Oracle and/or its affiliates 75