Scaling

Loom scales horizontally through agent replicas and vertically through provider capacity.

Agent Scaling

Docker Compose

make scale-coders N=3
make scale-reviewers N=2
make scale-qa N=2
# Or all at once:
make scale-agents CODERS=3 REVIEWERS=2 QA=2

Kubernetes

kubectl scale deployment loom-agent-coder -n loom --replicas=3

Provider Scaling

Add multiple providers to increase LLM throughput. Loom load-balances across healthy providers using weighted round-robin.

Database Scaling

• PgBouncer connection pooler is included by default (max 100 client connections, 20 pool size)
• PostgreSQL can be scaled to a managed service (RDS, Cloud SQL) by updating connection settings

Message Bus

NATS with JetStream handles inter-service messaging. For high-throughput:

• Enable NATS clustering for HA
• Increase JetStream storage limits
• Configure consumer concurrency per agent type

Observability at Scale

All services export metrics to Prometheus via OpenTelemetry:

• Monitor agent utilization to right-size replicas
• Track provider latency to identify bottlenecks
• Use Grafana dashboards for real-time visibility