agent-deployment

name	Agent Deployment
description	Production deployment workflow for agentic systems. Directs to RAG for implementation.

Agent Deployment Workflow

Deployment Decision Framework

Framework	Primary Option	Alternative	RAG Query
ADK	Agent Engine (Vertex AI)	Cloud Run, GKE	"ADK deployment agent engine"
OpenAI	Any Python hosting	Serverless, Docker	"openai agents deployment"
LangChain	LangServe, Cloud Run	Docker, K8s	"langchain langserve deployment"
LangGraph	LangGraph Platform	Cloud Run	"langgraph platform deployment"
CrewAI	CrewAI Enterprise	Docker	"crewai deployment production"
Anthropic	Any Python hosting	Docker, Serverless	"anthropic agent deployment"

Pre-Deployment Checklist

Code Readiness

• All tests passing
• Error handling complete
• Logging configured
• Input validation in place
• Output guardrails active

Configuration

• Environment variables documented
• Secrets in secret manager (NOT in .env or code)
• Rate limiting configured
• Token limits set
• Timeout values appropriate

Security

• API key rotation plan
• Audit logging enabled
• PII handling documented
• Input sanitization active
• Output filtering configured

Monitoring

• Health check endpoint
• Metrics collection
• Alerting rules defined
• Log aggregation setup

Deployment Workflow

Step 1: Environment Configuration

RAG Query: mcp__agentic-rag__search("[framework] environment configuration", mode="explain")

Production differs from development:

• LOG_LEVEL: INFO (not DEBUG)
• TRACE_ENABLED: false (or sampling)
• Secrets: Use secret manager, not .env

Step 2: Containerization (if applicable)

RAG Query: mcp__agentic-rag__search("[framework] dockerfile", mode="build")

Step 3: Platform Deployment

RAG Query: mcp__agentic-rag__search("[framework] [platform] deployment", mode="explain")

Step 4: Monitoring Setup

RAG Query: mcp__agentic-rag__search("[framework] monitoring observability", mode="explain")

Key Production Metrics

Metric	Alert Threshold	Why It Matters
Latency p95	> 5s	User experience
Error rate	> 1%	Reliability
Token usage	Spike > 200%	Cost control
Tool failures	> 5%	Agent effectiveness
Routing accuracy	< 90%	Multi-agent health

Security Considerations

Input Validation

• Sanitize user input before passing to agent
• Limit input length
• Filter known attack patterns

RAG Query: mcp__agentic-rag__search("agent input validation security", mode="explain")

Output Guardrails

• Filter sensitive information
• Prevent prompt leakage
• Validate tool outputs

RAG Query: mcp__agentic-rag__search("agent guardrails output filtering", mode="explain")

Secret Management

• Never hardcode API keys
• Use platform secret managers (GCP Secret Manager, AWS Secrets Manager, etc.)
• Rotate keys regularly

RAG Query: mcp__agentic-rag__search("[framework] secret management", mode="explain")

Scaling Considerations

Concern	Solution	RAG Query
Cold starts	Keep warm instances	"[framework] cold start"
Concurrent requests	Queue + workers	"[framework] scaling"
Token limits	Request batching	"[framework] rate limiting"
State persistence	External store	"[framework] state persistence"