On-Premises Deployment

CortexDB can run entirely on your infrastructure with no external dependencies. This guide covers fully air-gapped deployments.

Architecture

An on-prem CortexDB deployment consists of:

1. CortexDB cluster -- 3+ nodes running the CortexDB binary
2. Local LLM (optional) -- Ollama or vLLM for entity extraction
3. Local embeddings (optional) -- Local embedding model for vector search

No data leaves your network.

Air-Gapped Deployment

1. Load Docker Images

Transfer the CortexDB and Ollama images to your environment:

# On a machine with internet access
docker pull cortexdb/cortexdb:latest
docker pull ollama/ollama:latest
docker save cortexdb/cortexdb:latest | gzip > cortexdb.tar.gz
docker save ollama/ollama:latest | gzip > ollama.tar.gz

# Transfer files to air-gapped environment

# On the air-gapped machine
docker load < cortexdb.tar.gz
docker load < ollama.tar.gz

2. Load Models

# Pull models on an internet-connected machine
ollama pull llama3.2
ollama pull nomic-embed-text

# Export models
# Copy ~/.ollama/models to the air-gapped environment

3. Deploy

version: "3.8"

services:
  cortexdb:
    image: cortexdb/cortexdb:latest
    environment:
      CORTEX_LLM_PROVIDER: ollama
      CORTEX_LLM_BASE_URL: http://ollama:11434
      CORTEX_LLM_MODEL: llama3.2
      CORTEX_EMBEDDING_PROVIDER: ollama
      CORTEX_EMBEDDING_BASE_URL: http://ollama:11434
      CORTEX_EMBEDDING_MODEL: nomic-embed-text
      CORTEX_ENCRYPTION_AT_REST: "true"
      CORTEX_ENCRYPTION_KEY: "${CORTEX_ENCRYPTION_KEY}"
      CORTEX_API_KEY: "${CORTEX_API_KEY}"
    volumes:
      - cortexdb-data:/data
    ports:
      - "8080:8080"

  ollama:
    image: ollama/ollama:latest
    volumes:
      - ollama-models:/root/.ollama

volumes:
  cortexdb-data:
  ollama-models:

No-LLM Mode

CortexDB can operate without any LLM. Entity extraction and graph features are disabled, but remember/recall with keyword and vector search still works:

CORTEX_EXTRACTION_ENABLED=false
CORTEX_EMBEDDING_PROVIDER=local  # Uses built-in embedding model

Hardware Requirements

| Component | Minimum | Recommended | |---|---|---| | CortexDB node | 4 CPU, 4 GB RAM, 50 GB SSD | 8 CPU, 16 GB RAM, 200 GB NVMe | | Ollama (LLM) | 8 CPU, 16 GB RAM | GPU with 24 GB VRAM | | Ollama (embed only) | 2 CPU, 4 GB RAM | 4 CPU, 8 GB RAM |

Support

Enterprise on-prem deployments include:

• Dedicated support channel
• Deployment assistance
• Custom integration development
• SLA guarantees

Contact [email protected] for details.