Production Deployment

Docker Compose (Recommended)

The project includes a production-ready Docker Compose configuration:


docker compose -f docker-compose.prod.yml up -d

This starts three services:

Typesense — search engine with persistent data volume
Redis — persistent queue backend with AOF
Proxy — the tsproxy API server

Environment Variables

Create a .env file:


TYPESENSE_API_KEY=your-production-api-key
PROXY_API_KEY=your-ingest-secret
PROXY_PORT=3000
CACHE_TTL=60
CACHE_MAX_SIZE=1000
QUEUE_CONCURRENCY=5
QUEUE_MAX_SIZE=10000

Health Checks

All services include health checks. The proxy won’t start until Typesense and Redis are healthy:


curl http://localhost:3000/api/health


{
  "status": "healthy",
  "timestamp": "2026-04-03T12:00:00.000Z",
  "proxy": { "status": "ok" },
  "typesense": { "status": "ok", "host": "http://typesense:8108" },
  "redis": { "status": "ok", "host": "redis:6379" }
}

Building the Docker Image


docker build -t tsproxy .

The multi-stage Dockerfile:

Installs dependencies (pnpm)
Builds @tsproxy/js and @tsproxy/api
Produces a minimal Node.js 24 runtime image

Standalone Deployment

If you prefer running without Docker:


# Build
pnpm --filter @tsproxy/api build
 
# Start
TYPESENSE_API_KEY=your-key \
TYPESENSE_HOST=your-typesense-host \
REDIS_HOST=your-redis-host \
node packages/api/dist/server.js

Or use the CLI:


pnpm --filter @tsproxy/api start

Scaling

The proxy is stateless (cache is in-memory per instance, queue is in Redis)
Run multiple proxy instances behind a load balancer
All instances share the same Redis queue for ingestion
Each instance has its own LRU search cache