GroundX On-Prem provides a self-hostable, Kubernetes-deployable infrastructure for advanced Retrieval Augmented Generation (RAG) capabilities, including document parsing, secure storage, and semantic search. It is designed for enterprises requiring isolated, air-gapped environments and offers superior performance on complex documents compared to many popular RAG tools.

How It Works

GroundX On-Prem comprises three core services: Ingest, Store, and Search. The Ingest service utilizes a fine-tuned vision model to understand and convert diverse document formats into LLM-queryable representations. GroundX Store provides encrypted storage for source files, semantic objects, and vectors. The Search service, built on OpenSearch, combines text and vector search with a custom re-ranker model for precise retrieval. This multi-modal approach is designed for enterprise-grade accuracy and scalability.

Quick Start & Requirements

• Installation: Deploy via Helm to an existing Kubernetes cluster (v1.18+) or provision new infrastructure on AWS using provided Terraform scripts.
• Prerequisites: bash (v4.0+), terraform (setup docs), kubectl (setup docs). For AWS provisioning: AWS CLI (setup docs).
• Compute: Requires x86_64 architecture. GPU deployments necessitate NVIDIA GPUs with CUDA 12+. Resource requirements vary significantly by node group (CPU-only, CPU-memory, GPU-layout, GPU-ranker, GPU-summary), with recommended total resources including 40GB disk, 6 CPU cores, and 12GB RAM for basic CPU-only nodes, scaling up to 150GB disk, 11 CPU cores, and 48GB GPU memory for ranker nodes.
• Configuration: Requires creating an operator/env.tfvars file with admin credentials and potentially other configurations.
• Links: Setup Docs, AWS Deployment Guide, Online Testing Tool

Highlighted Details

• Designed for air-gapped environments with no external dependencies during operation.
• Leverages a custom OpenSearch configuration and fine-tuned re-ranker for enhanced search accuracy.
• Supports fine-tuning of the vision model for specific document sets.
• Offers SDKs for Python and TypeScript for seamless integration.

Maintenance & Community

• The project is in "Open Beta" and encourages feedback.
• Contact information for white-glove support and closed-source versions is provided.

Licensing & Compatibility

• The repository's license is not explicitly stated in the README.

Limitations & Caveats

• Terraform-based infrastructure creation is currently limited to AWS. Deployment to other Kubernetes clusters requires significant expertise.
• The project is in Open Beta, indicating potential for instability or breaking changes.
• Pods for architectures other than x86_64 (e.g., arm64) are available only upon customer request.