Skip to content
Hybrid GraphRAG Wiki & Architecture

Installation & Deployment Guide

Deployment Architecture

The complete system deploys as an interconnected multi-container environment via Docker Compose. The default stack includes:

  • PostgreSQL 16 (pgvector)
  • Qdrant Vector Database
  • Memgraph MAGE In-Memory Knowledge Graph and Memgraph Lab
  • RustFS S3-Compatible Object Store
  • Langfuse v4 Full Observability Stack (Web, Worker, ClickHouse, Redis, MinIO)
  • DOCX Regulatory Document Service
  • FastAPI Backend & FastMCP Server (hybrid-rag-api)
  • GxPChat Unified Next.js Application (hybrid-rag-gxpchat)
  • System Documentation Wiki (hybrid-rag-wiki)

Step-by-Step Installation Procedure

Step 1: Clone Repository and Enter Workspace

Terminal window
git clone https://github.com/saram-io/hybrid-graphrag.git
cd hybrid-graphrag

Step 2: Configure Environment Variables

Copy the template configuration file:

Terminal window
cp .env.example .env

Open .env in an editor and configure model credentials. If using DeepSeek foundation models:

Terminal window
DEEPSEEK_API_KEY=your_deepseek_api_key_here
DEEPSEEK_BASE_URL=https://api.deepseek.com
LLM_MODEL=deepseek-v4-flash

If using local Ollama models on the host machine:

Terminal window
EMBEDDING_PROVIDER=ollama
EMBEDDING_MODEL=qwen3-embedding:8b
EMBEDDING_DIMENSION=4096
OLLAMA_BASE_URL=http://host.docker.internal:11434

Step 3: Build and Launch Docker Containers

Execute Docker Compose in detached mode:

Terminal window
docker compose up -d --build

Verify that all containers reach a healthy state:

Terminal window
docker compose ps

Expected container states:

  • hybrid-rag-postgres: Up (healthy)
  • hybrid-rag-qdrant: Up (healthy)
  • hybrid-rag-memgraph: Up (healthy)
  • hybrid-rag-memgraph-lab: Up
  • hybrid-rag-rustfs: Up (healthy)
  • hybrid-rag-langfuse: Up (healthy)
  • hybrid-rag-clickhouse: Up (healthy)
  • hybrid-rag-docx-service: Up (healthy)
  • hybrid-rag-api: Up (healthy)
  • hybrid-rag-gxpchat: Up (healthy)
  • hybrid-rag-wiki: Up (healthy)

Step 4: Execute Life Sciences Domain Data Seeding

The platform includes an automated synthetic seeding pipeline that populates the three database engines with 505 controlled documents, 202 enterprise assets, 8 multi-agent squad scenarios, and backup/DR records:

Terminal window
docker compose exec api python -m src.seed.seed_all

The seed command automatically:

  1. Provisions PostgreSQL tables, constraints, and 21 CFR Part 11 triggers.
  2. Ingests SOPs, URS, FS, DS, and Test Protocols into PostgreSQL and calculates chunk hashes.
  3. Generates dense and sparse vector embeddings and loads them into Qdrant collections.
  4. Builds the Labeled Property Graph (LPG) inside Memgraph with multi-hop traceability edges.
  5. Populates the append-only backup and restoration qualification ledger.
  6. Stages sample binary documents in RustFS S3 buckets.

Step 5: Verify Web Applications and Services

Open a web browser and confirm access to the active service endpoints:

  • System Documentation Wiki: http://localhost:4321
  • GxPChat Consolidated Application Suite: http://localhost:3010 (or http://localhost:3002)
  • FastAPI OpenAPI Documentation: http://localhost:8090/docs
  • Memgraph Lab: http://localhost:3006
  • Qdrant Vector Dashboard: http://localhost:6335/dashboard
  • RustFS S3 Console: http://localhost:9121
  • Langfuse Observability: http://localhost:3001

System Documentation Wiki Portal

Step 6: Standalone Profile Execution (Optional)

If individual standalone micro-frontends are required on isolated ports:

Terminal window
docker compose --profile standalone up -d --build

This starts:

  • QMS Portal on port 3008
  • Asset Management App on port 3009
  • IT Help Desk App on port 3011
  • IT Validation App on port 3013