proffesor-for-testing__agentic-qe/docs/agentic-flow-setup/PRACTICAL_EXAMPLES.md

Agentic Flow - Practical Examples for AQE Project

🎯 Real-World Usage Examples

Example 1: Test Generation with Multi-Model Router

Scenario: Generate comprehensive test suite for a TypeScript module with cost optimization.

# Using DeepSeek Chat V3 (98% cost savings vs Claude Sonnet 4.5)
agentic-flow --agent qe-test-generator \
  --task "Generate comprehensive test suite for src/core/agents/BaseAgent.ts with 90% coverage" \
  --provider openrouter \
  --model "deepseek/deepseek-chat-v3.1:free" \
  --output json \
  --stream

# With automatic optimization
agentic-flow --agent qe-test-generator \
  --task "Generate comprehensive test suite for src/core/agents/BaseAgent.ts with 90% coverage" \
  --optimize --priority cost

Expected Output: Test suite with unit tests, integration tests, edge cases, and mocks.

Cost: ~$0.0002 per 1K tokens (vs ~$0.003 for Claude Sonnet)

Example 2: Multi-Agent Swarm with QUIC Transport

Scenario: Run multiple agents concurrently for end-to-end feature development.

# Terminal 1: Start QUIC server
agentic-flow quic --port 4433

# Terminal 2: Run multi-agent swarm
# Agent 1: Research and planning
agentic-flow --agent researcher \
  --task "Research best practices for authentication middleware" \
  --transport quic --optimize --priority speed &

# Agent 2: Implementation
agentic-flow --agent coder \
  --task "Implement JWT authentication middleware in TypeScript" \
  --transport quic --optimize --priority balanced &

# Agent 3: Testing
agentic-flow --agent qe-test-generator \
  --task "Create tests for JWT authentication middleware" \
  --transport quic --optimize --priority cost &

# Agent 4: Review
agentic-flow --agent reviewer \
  --task "Review JWT middleware security and code quality" \
  --transport quic --optimize --priority quality &

# Wait for all agents to complete
wait

Benefits:

• 50-70% faster communication vs HTTP
• Stream multiplexing: All 4 agents communicate without blocking
• Cost optimized: Each agent uses optimal model for its task type

Example 3: Coverage Analysis with Agent Booster

Scenario: Analyze test coverage and identify gaps using optimized models.

# Auto-select best model for analysis task
agentic-flow --agent qe-coverage-analyzer \
  --task "Analyze test coverage for agentic-qe/ directory and identify gaps" \
  --optimize --priority balanced \
  --output json

# With cost constraint (max $0.001 per task)
agentic-flow --agent qe-coverage-analyzer \
  --task "Analyze test coverage for agentic-qe/ directory and identify gaps" \
  --optimize --max-cost 0.001

Output: JSON report with coverage metrics, uncovered lines, and recommendations.

Example 4: Security Scanning with Quality Priority

Scenario: Comprehensive security audit requiring high-quality analysis.

# Use premium model for security-critical task
agentic-flow --agent qe-security-scanner \
  --task "Perform security audit of authentication and authorization code" \
  --optimize --priority quality \
  --verbose

# Fallback if quality models unavailable
agentic-flow --agent qe-security-scanner \
  --task "Perform security audit of authentication and authorization code" \
  --model "openai/gpt-4-turbo" \
  --verbose

Output: Security vulnerabilities, CVE matches, SAST/DAST findings, recommendations.

Example 5: Claude Code Integration via Proxy

Scenario: Use OpenRouter models directly in Claude Code CLI.

# Terminal 1: Start proxy with OpenRouter
agentic-flow proxy --provider openrouter --port 3000

# Terminal 2: Configure and use Claude Code
export ANTHROPIC_BASE_URL="http://localhost:3000"
export ANTHROPIC_API_KEY="sk-ant-proxy-dummy-key"

# Now use Claude Code normally - it will route through OpenRouter
claude-code

# Inside Claude Code, all operations use OpenRouter models
# Example: "Write a REST API for user management"

Benefits:

• 85-99% cost savings on all Claude Code operations
• Access to 100+ OpenRouter models
• No code changes to Claude Code
• Leaderboard tracking on OpenRouter

Example 6: Performance Testing with QUIC

Scenario: Load testing with high-frequency agent coordination.

# Start QUIC server for low-latency coordination
agentic-flow quic --port 4433 &

# Run performance test agent
agentic-flow --agent qe-performance-tester \
  --task "Load test API endpoints with 1000 concurrent requests" \
  --transport quic \
  --optimize --priority speed \
  --stream

# Multiple performance agents in parallel
for i in {1..5}; do
  agentic-flow --agent qe-performance-tester \
    --task "Load test endpoint /api/v1/users with 200 concurrent requests" \
    --transport quic \
    --optimize --priority speed &
done
wait

QUIC Benefits:

• 0-RTT reconnection: Agents reconnect instantly
• Connection migration: Survives network interruptions
• 100+ concurrent streams: All agents coordinate in real-time

Example 7: Documentation Generation (Cost Optimized)

Scenario: Generate API documentation for multiple modules.

# Use cheapest model for documentation task
agentic-flow --agent api-docs \
  --task "Generate OpenAPI documentation for REST API endpoints" \
  --optimize --priority cost \
  --output md

# Batch processing multiple files
for file in src/api/*.ts; do
  agentic-flow --agent api-docs \
    --task "Generate API docs for $file" \
    --optimize --priority cost \
    --output md > "docs/api/$(basename $file .ts).md" &
done
wait

Cost: ~$0.0001 per 1K tokens (Llama 3.1 8B or DeepSeek Chat V3)

Example 8: Privacy-Focused Code Review (Local Models)

Scenario: Review sensitive code using local ONNX models (no cloud).

# Use local ONNX Phi-4 model (no API calls)
agentic-flow --agent reviewer \
  --task "Review security-sensitive authentication code" \
  --optimize --priority privacy \
  --provider onnx

# Alternative: explicit ONNX provider
agentic-flow --agent reviewer \
  --task "Review security-sensitive authentication code" \
  --provider onnx

Benefits:

• 100% private: No data leaves your machine
• No API costs: Free local inference
• Fast: ~2-5 seconds for code review

Example 9: Complete Feature Development Pipeline

Scenario: End-to-end feature with all optimizations enabled.

#!/bin/bash
# feature-pipeline.sh - Complete feature development with agentic-flow

# Start QUIC server for coordination
agentic-flow quic --port 4433 &
QUIC_PID=$!

# Phase 1: Research and Planning (speed optimized)
echo "Phase 1: Research"
agentic-flow --agent researcher \
  --task "Research user authentication patterns and security best practices" \
  --transport quic --optimize --priority speed \
  --output json > research.json

# Phase 2: Architecture Design (balanced)
echo "Phase 2: Architecture"
agentic-flow --agent system-architect \
  --task "Design authentication system architecture based on research.json" \
  --transport quic --optimize --priority balanced \
  --output json > architecture.json

# Phase 3: Implementation (quality for production code)
echo "Phase 3: Implementation"
agentic-flow --agent coder \
  --task "Implement authentication system based on architecture.json" \
  --transport quic --optimize --priority quality \
  --output text > implementation.log

# Phase 4: Testing (cost optimized)
echo "Phase 4: Testing"
agentic-flow --agent qe-test-generator \
  --task "Generate comprehensive test suite for authentication system" \
  --transport quic --optimize --priority cost \
  --output json > tests.json

agentic-flow --agent qe-test-executor \
  --task "Execute test suite and report results" \
  --transport quic --optimize --priority cost \
  --output json > test-results.json

# Phase 5: Security & Quality (quality for critical analysis)
echo "Phase 5: Security & Quality"
agentic-flow --agent qe-security-scanner \
  --task "Perform security audit of authentication implementation" \
  --transport quic --optimize --priority quality \
  --output json > security-report.json

agentic-flow --agent reviewer \
  --task "Perform code review focusing on maintainability and best practices" \
  --transport quic --optimize --priority quality \
  --output json > review-report.json

# Phase 6: Documentation (cost optimized)
echo "Phase 6: Documentation"
agentic-flow --agent api-docs \
  --task "Generate comprehensive API documentation" \
  --transport quic --optimize --priority cost \
  --output md > docs/authentication-api.md

# Cleanup
kill $QUIC_PID

echo "✅ Complete pipeline finished!"
echo "📊 Generated artifacts:"
ls -lh research.json architecture.json implementation.log tests.json test-results.json security-report.json review-report.json docs/authentication-api.md

Example 10: Monitoring and Observability

Scenario: Monitor multi-agent execution with real-time metrics.

# Start QUIC server with verbose logging
agentic-flow quic --port 4433 --verbose &

# Run agent with streaming output
agentic-flow --agent coder \
  --task "Build REST API with Express and TypeScript" \
  --transport quic \
  --optimize --priority balanced \
  --stream \
  --verbose

# Monitor QUIC metrics
watch -n 1 'lsof -i :4433 | tail -n +2 | wc -l'

🎯 Integration with AQE Project

Quick Start Script

Create scripts/agentic-flow-setup.sh:

#!/bin/bash
# Setup agentic-flow for AQE project

# Load environment variables
source ~/.bashrc

# Configure agentic-flow
echo "🔧 Configuring agentic-flow..."
agentic-flow config set PROVIDER openrouter
agentic-flow config set COMPLETION_MODEL "deepseek/deepseek-chat-v3.1:free"

# Start QUIC server
echo "🚀 Starting QUIC server..."
agentic-flow quic --port 4433 &
QUIC_PID=$!

# Wait for QUIC server
sleep 2

# Test configuration
echo "✅ Testing configuration..."
agentic-flow --agent coder \
  --task "Write a hello world function" \
  --transport quic \
  --optimize --priority balanced \
  --stream

echo "✅ Setup complete!"
echo "QUIC server running on port 4433 (PID: $QUIC_PID)"
echo ""
echo "To stop QUIC server: kill $QUIC_PID"

Make executable and run:

chmod +x scripts/agentic-flow-setup.sh
./scripts/agentic-flow-setup.sh

📊 Performance Comparison

Configuration	Latency	Cost per 1M tokens	Throughput
Claude Sonnet 4.5 (HTTP)	100-200ms	$15.00	1x
OpenRouter + HTTP	80-150ms	$0.30-$2.00	1.2x
OpenRouter + QUIC	30-70ms	$0.30-$2.00	2.8-4.4x
OpenRouter + QUIC + Optimize (cost)	30-70ms	$0.03-$0.20	2.8-4.4x

🎓 Best Practices

1. Use QUIC for multi-agent coordination: 50-70% faster than HTTP
2. Use optimization for cost savings: 85-98% cheaper than Claude
3. Match priority to task type:
   • quality: Production code, security audits
   • balanced: General development (default)
   • cost: Documentation, simple tasks
   • speed: Research, quick queries
   • privacy: Sensitive data
4. Use streaming for long tasks: See progress in real-time
5. Use JSON output for automation: Easier to parse and process
6. Monitor costs: Use --max-cost flag for budget control
7. Leverage free models: DeepSeek, Llama 3.1 8B are free on OpenRouter

🔗 Related Resources

• Configuration Guide
• Agentic Flow GitHub
• OpenRouter Models
• Model Capabilities Benchmark