Part of GPT‑RAG
Part of the GPT-RAG solution.
The GPT-RAG Orchestrator service is an agentic orchestration layer built on Azure AI Foundry Agent Service and the Semantic Kernel framework. It enables agent-based RAG workflows by coordinating multiple specialized agents—each with a defined role—to collaboratively generate accurate, context-aware responses for complex user queries.
The orchestrator uses Azure AI Foundry Agent Service for single-agent and connected-agent flows, leveraging its managed runtime for agent lifecycle, state, and tool orchestration. For multi-agent scenarios, it integrates the Semantic Kernel Agent Framework to compose and coordinate specialized agents collaborating on tasks. Custom agent strategies allow developers to plug domain-specific logic without modifying core orchestration code.
Developers can extend the orchestrator by creating a new subclass of BaseAgentStrategy, implementing the required initiate_agent_flow method and any additional helpers, then registering it in AgentStrategyFactory.get_strategy under a unique key. The base class provides shared logic (e.g., prompt loading via _read_prompt, credential setup) so extensions focus only on custom behavior.
Provision the infrastructure first by following the GPT-RAG repository instructions GPT-RAG. This ensures all required Azure resources (e.g., Container App, Storage, AI Search) are in place before deploying the web application.
Click to view software prerequisites
The machine used to customize and or deploy the service should have:
- Azure CLI: Install Azure CLI
- Azure Developer CLI (optional, if using azd): Install azd
- Git: Download Git
- Python 3.12: Download Python 3.12
- Docker CLI: Install Docker
- VS Code (recommended): Download VS Code
Click to view permissions requirements
To customize the service, your user should have the following roles:
| Resource | Role | Description |
|---|---|---|
| App Configuration Store | App Configuration Data Owner | Full control over configuration settings |
| Container Registry | AcrPush | Push and pull container images |
| Key Vault | Key Vault Contributor | Manage Key Vault Secrets |
| AI Search Service | Search Service Contributor | Create or update search service components |
| AI Search Service | Search Index Data Contributor | Read and write index data |
| Storage Account | Storage Blob Data Contributor | Read and write blob data |
| AI Foundry Project | Azure AI Project User | Access and work with the AI Foundry project |
| Cosmos DB | Cosmos DB Built-in Data Contributor | Read and write documents in Cosmos DB |
To deploy the service, assign these roles to your user or service principal:
| Resource | Role | Description |
|---|---|---|
| App Configuration Store | App Configuration Data Reader | Read config |
| Container Registry | AcrPush | Push images |
| Azure Container App | Azure Container Apps Contributor | Manage Container Apps |
Ensure the deployment identity has these roles at the correct scope (subscription or resource group).
Make sure you're logged in to Azure before anything else:
az loginClone this repository.
Just run:
azd env refresh
azd deploy Important
Make sure you use the same subscription, resource group, environment name, and location from azd provision.
You need to set the App Configuration endpoint and run the deploy script.
export APP_CONFIG_ENDPOINT="https://<your-app-config-name>.azconfig.io"
./scripts/deploy.sh$env:APP_CONFIG_ENDPOINT = "https://<your-app-config-name>.azconfig.io"
.\scripts\deploy.ps1Note
If you want to use the GPT-RAG original version, simply use the v1.0.0 release from the GitHub repository.
We appreciate contributions! See CONTRIBUTING.md for guidelines on the Contributor License Agreement (CLA), code of conduct, and submitting pull requests.
This project may contain trademarks or logos. Authorized use of Microsoft trademarks or logos must follow Microsoft’s Trademark & Brand Guidelines. Modified versions must not imply sponsorship or cause confusion. Third-party trademarks are subject to their own policies.