This directory contains example integrations and extensions for NVIDIA RAG.
| Example | Description | Documentation |
|---|---|---|
| rag_react_agent | Integration with NeMo Agent Toolkit (NAT) providing RAG query and search capabilities for agent workflows | README |
| nvidia_rag_mcp | MCP (Model Context Protocol) server and client for exposing NVIDIA RAG capabilities to MCP-compatible applications | Documentation |
| rag_event_ingest | Automated document ingestion from object storage (MinIO) via Kafka | Notebook |
| google-cloud-netapp-volumes-data-ingestor | Helm chart for deploying the GCNV data ingestor with PVC-backed storage and configurable runtime settings | README |
This plugin integrates NVIDIA RAG with NeMo Agent Toolkit, enabling intelligent agents to use RAG tools for document retrieval and question answering. It demonstrates:
- Creating custom NAT tools that wrap NVIDIA RAG functionality
- Using the React Agent workflow for intelligent tool selection
See the rag_react_agent README for setup and usage instructions.
This example provides an MCP server and client that exposes NVIDIA RAG and Ingestor capabilities as MCP tools. It supports multiple transport modes (SSE, streamable HTTP, stdio) and enables MCP-compatible applications to:
- Generate answers using the RAG pipeline
- Search the vector database for relevant documents
- Manage collections and documents in the vector database
See the MCP documentation for detailed setup and usage instructions.
This example deploys an event-driven ingestion pipeline that monitors MinIO object storage for new file uploads via Kafka events. Documents are automatically indexed through the RAG Ingestor and become queryable through the RAG Agent.
Components:
- kafka_consumer/ - Event-driven consumer that routes files to RAG based on file type
- deploy/ - Docker Compose for Kafka, MinIO, and the consumer
- data/ - Sample documents for testing
See the notebook for step-by-step deployment and testing.
This example packages a GCNV data ingestor deployment as a reusable Helm chart. It is intended for Kubernetes environments where application state and source data are mounted from PVCs, including NetApp Google Cloud NetApp Volumes-backed storage.
The chart supports configurable image settings, PVC creation or reuse, health probes, service exposure, and runtime environment overrides for connecting to an NVIDIA ingestor endpoint.
See the google-cloud-netapp-volumes-data-ingestor README for prerequisites, installation, and configuration details.