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

Agentic Flow Configuration Guide

📋 Current Setup Status

✅ Environment Variables: Configured in ~/.bashrc

• ANTHROPIC_API_KEY: Set
• OPENROUTER_API_KEY: Set

✅ Installation: agentic-flow v1.6.4 installed globally

🎯 Three Key Features Setup

1. Multi-Model Router (OpenRouter Integration)

The Multi-Model Router allows you to use 100+ models from OpenRouter instead of just Claude models, with 85-99% cost savings.

Configuration Methods

Method A: Environment Variables (Recommended for global use)

Add to ~/.bashrc (already done for API keys):

export USE_OPENROUTER=true
export COMPLETION_MODEL="deepseek/deepseek-chat-v3.1:free"  # or any OpenRouter model

Then reload:

source ~/.bashrc

Method B: CLI Flags (Recommended for per-task control)

# Use OpenRouter with specific provider
agentic-flow --agent coder --task "Create API" --provider openrouter

# Use specific OpenRouter model
agentic-flow --agent coder --task "Create API" --model "meta-llama/llama-3.1-8b-instruct"

# Use optimization to auto-select best model
agentic-flow --agent coder --task "Create API" --optimize --priority balanced

Method C: Configuration File

# Set default provider
agentic-flow config set PROVIDER openrouter
agentic-flow config set COMPLETION_MODEL "deepseek/deepseek-chat-v3.1:free"

# Verify configuration
agentic-flow config list

Best Free OpenRouter Models

# For reasoning tasks (95s/task, RFC validation)
--model "deepseek/deepseek-r1-0528:free"

# For coding tasks (21-103s/task, enterprise-grade)
--model "deepseek/deepseek-chat-v3.1:free"

# For fast coding (4.4s/task, versatile)
--model "meta-llama/llama-3.3-8b-instruct:free"

# For premium quality (10.7s/task)
--model "openai/gpt-4-turbo"

Proxy Mode for Claude Code Integration

This allows Claude Code itself to use OpenRouter models:

Terminal 1 - Start Proxy:

# Start proxy on default port 3000
agentic-flow proxy --provider openrouter

# Or with custom port
PROXY_PORT=8080 agentic-flow proxy --provider openrouter

Terminal 2 - Configure and Use Claude Code:

export ANTHROPIC_BASE_URL="http://localhost:3000"
export ANTHROPIC_API_KEY="sk-ant-proxy-dummy-key"
export OPENROUTER_API_KEY="sk-or-v1-xxxxx"  # Your real key

# Now Claude Code will route through OpenRouter
claude-code

2. QUIC Transport (Ultra-Low Latency)

QUIC provides 50-70% faster agent communication than traditional TCP/HTTP.

Benefits

• 0-RTT Connection: Instant reconnection without handshake delay
• Stream Multiplexing: 100+ concurrent messages without blocking
• Built-in TLS 1.3: Encrypted by default
• Connection Migration: Survives network changes (WiFi → cellular)
• Reduced Latency: Perfect for real-time agent coordination

Configuration Methods

Method A: Environment Variable (Global Default)

Add to ~/.bashrc:

export AGENTIC_FLOW_TRANSPORT=quic
export QUIC_PORT=4433

Then reload:

source ~/.bashrc

Method B: CLI Flag (Per-Task)

# Use QUIC transport for single task
agentic-flow --agent coder --task "Build API" --transport quic

# With OpenRouter
agentic-flow --agent coder --task "Build API" --transport quic --provider openrouter

Method C: Start QUIC Server

# Start QUIC server on default port 4433
agentic-flow quic --port 4433

# With custom certificates
agentic-flow quic --cert ./certs/cert.pem --key ./certs/key.pem

Programmatic Usage

import { QuicTransport } from 'agentic-flow/transport/quic';

const transport = new QuicTransport({
  host: 'localhost',
  port: 4433,
  maxConcurrentStreams: 100
});

await transport.connect();
await transport.send({
  type: 'task',
  data: {
    agent: 'coder',
    task: 'Build REST API'
  }
});

Use Cases

• Multi-agent swarm coordination (mesh/hierarchical topologies)
• High-frequency task distribution across worker agents
• Real-time state synchronization between agents
• Low-latency RPC for distributed agent systems

3. Agent Booster (Model Optimization)

Agent Booster automatically selects the best model for your task based on priorities, providing 85-98% cost savings.

Features

• Smart Selection: Agent-aware (coder needs quality ≥85, researcher flexible)
• 10+ Models: Claude, GPT-4o, Gemini, DeepSeek, Llama, ONNX local
• Zero Overhead: <5ms decision time, no API calls during optimization
• Cost Control: Set max budget per task

Configuration Methods

Method A: CLI Flags (Recommended)

# Auto-optimize with balanced priority (default)
agentic-flow --agent coder --task "Build API" --optimize

# Optimize for cost (85-98% cheaper)
agentic-flow --agent coder --task "Build API" --optimize --priority cost

# Optimize for quality
agentic-flow --agent reviewer --task "Security audit" --optimize --priority quality

# Optimize for speed
agentic-flow --agent researcher --task "Find examples" --optimize --priority speed

# Optimize with cost budget
agentic-flow --agent coder --task "Simple function" --optimize --max-cost 0.001

Method B: Short Flag

# Use -O instead of --optimize
agentic-flow --agent coder --task "Build API" -O --priority balanced

Optimization Priorities

Priority	Best For	Example Models	Use Case
quality	Production code, security	Claude Sonnet 4.5, GPT-4o	Code reviews, security audits
balanced	Most tasks (default)	DeepSeek R1, Gemini 2.5 Flash	General development, testing
cost	Simple tasks, bulk operations	DeepSeek Chat V3, Llama 3.1 8B	Documentation, simple scripts
speed	Research, quick queries	Gemini 2.5 Flash	Quick lookups, research
privacy	Sensitive data	ONNX Phi-4 (local)	Private/confidential work

Cost Savings Examples

# DeepSeek R1: 85% less than Claude Sonnet 4.5
agentic-flow --agent coder --task "Build API" --model "deepseek/deepseek-r1-0528:free"

# DeepSeek Chat V3: 98% less for simple tasks
agentic-flow --agent coder --task "Add docstring" --model "deepseek/deepseek-chat-v3.1:free"

# Llama 3.1 8B: 99% less for basic tasks
agentic-flow --agent researcher --task "Find examples" --model "meta-llama/llama-3.3-8b-instruct:free"

🚀 Recommended Setup for Your Project

Based on your agentic-qe-cf project, here's the optimal configuration:

Step 1: Add Environment Variables

Add these to your ~/.bashrc (in addition to existing API keys):

# Multi-Model Router (Already have API keys ✅)
export USE_OPENROUTER=true
export COMPLETION_MODEL="deepseek/deepseek-chat-v3.1:free"

# QUIC Transport (Ultra-fast agent coordination)
export AGENTIC_FLOW_TRANSPORT=quic
export QUIC_PORT=4433

# Model Optimization (Enable by default)
export AGENTIC_FLOW_OPTIMIZE=true
export AGENTIC_FLOW_PRIORITY=balanced

Then reload:

source ~/.bashrc

Step 2: Configure agentic-flow

# Set default provider
agentic-flow config set PROVIDER openrouter
agentic-flow config set COMPLETION_MODEL "deepseek/deepseek-chat-v3.1:free"

# Verify configuration
agentic-flow config list

Step 3: Test Each Feature

# Test Multi-Model Router
agentic-flow --agent coder --task "Write a hello world function" --provider openrouter

# Test QUIC Transport
agentic-flow --agent coder --task "Write a hello world function" --transport quic

# Test Agent Booster
agentic-flow --agent coder --task "Write a hello world function" --optimize --priority cost

# Test All Together
agentic-flow --agent coder --task "Write a hello world function" --provider openrouter --transport quic --optimize --priority balanced

📊 Performance Comparison

Configuration	Speed	Cost	Best For
Standard (Claude)	1x		High-quality production code
OpenRouter + QUIC	1.5-1.7x	$	Fast development, testing
OpenRouter + QUIC + Optimize (cost)	1.5-1.7x	$	Bulk operations, documentation
OpenRouter + QUIC + Optimize (balanced)	1.5-1.7x	$$	General development (recommended)

🎯 Integration with Your Project

For AQE Fleet Operations

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

# Initialize AQE fleet with optimized models
cd /workspaces/agentic-qe-cf
aqe init

# Run test generation with optimization
agentic-flow --agent qe-test-generator \
  --task "Generate tests for src/core/agents/BaseAgent.ts" \
  --provider openrouter \
  --transport quic \
  --optimize --priority balanced

For Swarm Coordination

# Multi-agent execution with QUIC transport
agentic-flow --agent coder --task "Build REST API" --transport quic --optimize &
agentic-flow --agent tester --task "Create test suite" --transport quic --optimize &
agentic-flow --agent reviewer --task "Code review" --transport quic --optimize &

🔧 Troubleshooting

Issue: QUIC connection fails

Solution 1: Check if port 4433 is available

lsof -i :4433

Solution 2: Use custom port

export QUIC_PORT=5433
agentic-flow quic --port 5433

Issue: OpenRouter API key not recognized

Solution: Verify key is loaded

source ~/.bashrc
echo $OPENROUTER_API_KEY

Issue: Model optimization not working

Solution: Ensure optimization flag is used

agentic-flow --agent coder --task "Build API" --optimize --priority balanced

📚 Additional Resources

• Agentic Flow GitHub
• OpenRouter Models
• QUIC Protocol
• Model Capabilities Benchmark

🎓 Next Steps

1. ✅ Verify environment variables are loaded
2. ✅ Configure agentic-flow defaults
3. ✅ Test each feature individually
4. ✅ Test combined features
5. ✅ Integrate with your AQE project workflows
6. ✅ Monitor performance and costs
7. ✅ Optimize based on usage patterns