Agent Lifecycle

Overview

Agents progress through predictable states to optimize compute costs and availability.

Deploy → [Start ↔ Stop] → Terminate
         ↑
      Resume

Key Benefits:

Pause anytime → Save 95% compute costs
State preserved → Resume instantly
Predictable billing → Per-second granularity

Lifecycle Flow

graph TD
    A["🚀 Deploy"] --> B["⏳ Provisioned"]
    B -->|"▶️ Start"| C["🟢 Running"]
    B -->|"Never started"| D["🟡 Paused"] 
    C -->|"⏸️ Stop"| D
    D -->|"▶️ Start/Resume"| C
    C -->|"🗑️ Terminate"| E["❌ Terminated"]
    D -->|"🗑️ Terminate"| E
    E -->|"🚀 Redeploy"| A
    
    classDef running fill:#90EE90
    classDef paused fill:#FFA500
    classDef terminated fill:#ff9999
    class C running
    class D paused
    class E terminated

State Details

State	Endpoint	Compute	Data	Use Case
🚀 Deploy	202 Accepted	Provisioning	Creating	Initial setup
🟢 Running	✅ Active	GPU/CPU/Mem	Live	Production
🟡 Paused	503 Unavailable	Storage only	Preserved	Cost savings
❌ Terminated	Gone	None	Backup (30d)	Cleanup

Management Commands

# Core lifecycle
moltghost agent start   my-agent     # ▶️  Resume (45s)
moltghost agent stop    my-agent     # ⏸️  Pause (-95% cost)
moltghost agent delete  my-agent     # 🗑️  Terminate

# Smart defaults
moltghost agent pause    my-agent    # Auto-stop if idle
moltghost agent resume   my-agent    # Smart start

# Bulk operations
moltghost agent stop dev-*           # Pause all dev agents
moltghost agent start prod-*         # Start production fleet

Cost Impact

Running  (A100 80GB): $2.50/hour
Paused             : $0.12/hour  ← 95% savings
Terminated         : $0.00/hour

Monthly Example:
- Always-on: $1,800
- Smart pause: $450  ← 75% savings

Auto-Policy:

moltghost agent set my-agent --auto-pause "15m idle"

Summary

Simple states, powerful control:

✅ Deploy → Create infrastructure
✅ Start/Stop → Control costs precisely
✅ Terminate → Clean resource release
✅ Data always safe → Pause/resume anytime

Lifecycle mastery = cost mastery.