Deployment

Sentinel is designed for flexible deployment across environments—from single-binary development setups to distributed Kubernetes clusters.

Deployment Philosophy

Sentinel follows a separation of concerns model:

Component	Responsibility
Sentinel proxy	Route traffic, call agents, circuit breaking
Agents	Security logic, custom processing
Process supervisor	Lifecycle management (systemd, Docker, K8s)

Sentinel intentionally does not manage agent lifecycles. Process supervision is a solved problem—systemd, Docker, and Kubernetes do it better than we could. This keeps the proxy lean and focused.

Deployment Tiers

┌─────────────────────────────────────────────────────────────────┐
│                     DEPLOYMENT OPTIONS                          │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  Development:        sentinel-stack                             │
│                      └── Single command, spawns everything      │
│                                                                 │
│  Production (VMs):   systemd with socket activation             │
│                      └── Independent services, proper isolation │
│                                                                 │
│  Cloud-native:       Kubernetes / Docker Compose                │
│                      └── Containers, sidecars, service mesh     │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Quick Comparison

Deployment	Best For	Agents	Complexity
sentinel-stack	Development, simple setups	Child processes	Minimal
systemd	Production VMs, bare metal	Socket-activated services	Low
Docker Compose	Local development, small prod	Sidecar containers	Medium
Kubernetes	Cloud-native, scale-out	Pods, service mesh	Higher

Agent Connectivity

Regardless of deployment model, agents connect via:

• Unix sockets — Local agents, lowest latency (~50-100µs)
• gRPC — Remote agents, scalable, polyglot (~100-500µs)

See Agent Transports for protocol details.

Documentation

Page	Description
Architecture	Deployment philosophy and agent lifecycle
sentinel-stack	All-in-one launcher for development
systemd	Production deployment with systemd
Docker Compose	Container-based local/small deployments
Kubernetes	Cloud-native deployment patterns
Service Mesh	Istio, Linkerd, and Consul Connect integration
Monitoring	Observability and health checks