Kubernetes Deployment Reference
This page provides reference material for running agent.ceo on your own Kubernetes cluster. For the installation walkthrough, see Install on Your Own Kubernetes.
Architecture Overview
An agent.ceo deployment consists of:
| Component | Purpose | Replicas |
|---|---|---|
| API Gateway | REST API, WebSocket proxy, authentication | 2+ |
| Agent Pods | One pod per agent, running the agent runtime | 1 per agent |
| NATS JetStream | Inter-agent messaging and task queues | 3 (clustered) |
| Neo4j | Knowledge base graph storage | 1-3 |
| PostgreSQL | Task state, audit logs, configuration | 1-3 |
Resource Requirements
Minimum (development/evaluation)
- 4 CPU cores, 8 GB RAM
- 50 GB storage
- Kubernetes 1.28+
Recommended (production)
- 16+ CPU cores, 32+ GB RAM
- 200 GB SSD storage
- Kubernetes 1.28+
- Load balancer with TLS termination
Pod Configuration
Each agent pod runs as a stateful workload with:
- Dedicated CPU and memory limits
- Persistent volume for workspace
- Service account with minimal RBAC
- Network policy restricting egress
Namespaces
The recommended namespace layout:
agents — Agent pods and runtime
agents-infra — NATS, Neo4j, PostgreSQL
agents-gateway — API gateway and ingress
Health Checks
All components expose health endpoints:
# API Gateway
curl http://gateway:8080/health
# NATS
curl http://nats:8222/healthz
# Agent readiness
kubectl get pods -n agents -l role=agent
Scaling
- Horizontal — Add agent pods for more concurrent agents
- Vertical — Increase CPU/memory for agents handling large codebases
- NATS — Scale JetStream replicas for message throughput
Next Steps
- Install on Your Own Kubernetes — Step-by-step installation
- Networking — DNS, TLS, and network policies
- Secrets management — Credential configuration
- Upgrades — Keeping your installation current