Oracle AI Optimizer and Toolkit

The Oracle AI Optimizer and Toolkit (the AI Optimizer) is an open-source platform for building, testing, and evaluating GenAI applications. Its agentic runtime grounds model responses at request time in authoritative private and sensitive data that remains outside the model's training knowledge and under your control in Oracle AI Database.

What You Can Do

• Run LLM-only and tool-enabled agents and flows.
• Ground responses in unstructured content by splitting and embedding documents for Vector Search.
• Query structured data in Oracle AI Database through the NL2SQL agent.
• Compare models, prompts, parameters, and retrieval settings in the Chatbot.
• Evaluate configurations against generated or existing Q&A datasets in the Testbed.
• Apply identity-aware database authorization with Oracle Deep Data Security.
• Connect external applications through the REST API or MCP endpoint.

Technology

The AI Optimizer brings together established open-source components^*:

• FastAPI for the REST API

• FastMCP for MCP tools, prompts, and resources

• LangGraph and AgentSpec for agent and flow orchestration

• Docling for document parsing

• LiteLLM for model-provider integration

• Giskard for evaluation

• OpenTelemetry for observability

*Project names and logos are shown for identification purposes only. Unless expressly stated, their use does not imply affiliation with, sponsorship by, or endorsement by Oracle or the respective project owners.

Architecture

The AI Optimizer Server hosts the REST API, agent and flow runtime, and MCP endpoint. The optional AI Optimizer Client provides a web interface for configuring and exercising the Server through its REST API. Other applications can integrate with the Server through the REST API or MCP endpoint.

Requirements

The Server requires an Oracle AI Database connection for application configuration and persistence. The routes you run determine the remaining requirements:

• A language model runs agents and flows.
• An embedding model creates vectors for RAG and Vector Search.
• SQLcl must be installed where the Server runs and available on PATH to use NL2SQL.

The client can run alongside the Server or connect to a separately deployed Server. All components can be colocated or distributed across local, on-premises, and cloud environments.

Get Started

• Follow the Walkthrough for a guided local environment and introduction to the main features.
• Use the Quick Start to choose a bare-metal, container, OCI, or Kubernetes deployment.
• Explore the AI Optimizer Client to configure and exercise the included workbench.
• Explore the AI Optimizer Server to integrate through REST or MCP.