Self-Hosted Deployment

Run QANATIX on your own infrastructure. Full control, air-gapped support, data never leaves your network.

Prerequisites

Docker and Docker Compose v2
4 GB RAM minimum (8 GB recommended)
GPU recommended for large datasets (see Embedding Model for hardware guidance)

Docker Compose

See Installation for the full docker-compose.yml.

Production checklist

Before going live:

Set strong passwords for Postgres and Redis
Set API_KEY_SALT to a random 32+ character string
Set CONNECTOR_ENCRYPTION_KEY for database connector credential encryption
Set DEPLOYMENT_MODE=self_hosted
Set up backup for Postgres volumes
Put a reverse proxy (nginx/Caddy) in front for TLS
Set CORS_ORIGINS to your domain(s)

Scaling

Component	Scaling strategy
API	Horizontal — run multiple `qanatix-api` containers behind a load balancer
Worker	Horizontal — run multiple `qanatix-worker` containers (SAQ distributes jobs)
Postgres	Vertical — single primary, add read replicas for search queries
Redis	Single instance is sufficient for most workloads
Qdrant	Horizontal — Qdrant supports distributed mode

BGE-M3 runs locally — no API keys needed. It produces dense, sparse, and ColBERT vectors simultaneously for maximum search quality. CPU works fine for small-medium datasets; GPU recommended for >100K entities.

GPU	VRAM	Throughput	Cost
NVIDIA T4	16 GB	~100 embeddings/sec	Budget option
NVIDIA L4	24 GB	~300 embeddings/sec	Best value
NVIDIA A10G	24 GB	~300 embeddings/sec	Cloud standard

Air-gapped

For classified environments with no internet:

Pre-download the BGE-M3 model weights on an internet-connected machine
Pre-pull all Docker images to an internal registry
No outbound network required after setup

See Air-gapped for detailed setup.

See Configuration for all environment variables.

Self-Hosted