proffesor-for-testing__agentic-qe/examples/configs

Provider Configuration Examples

This directory contains example configurations for different deployment scenarios.

Configuration Files

1. local-first.yaml

Use Case: Development, cost-sensitive environments, privacy-focused

Strategy:

• Primary: Ollama (local inference, 100% free)
• Fallback: OpenRouter (cost-effective cloud models)

Best For:

• Local development
• Privacy-sensitive projects
• Budget-constrained teams
• Offline-capable deployments

Cost: ~$0-5/day (mostly free with Ollama)

---

2. hosted-only.yaml

Use Case: Production environments, CI/CD pipelines, enterprise teams

Strategy:

• Primary: Groq (fast, generous free tier)
• Secondary: OpenRouter (multi-model routing)
• Tertiary: Claude API (premium quality)

Best For:

• Production deployments
• CI/CD integration
• Teams without local GPU infrastructure
• High-reliability requirements

Cost: ~$10-50/day (depends on usage)

---

3. free-tier.yaml

Use Case: Open-source projects, learning, personal use

Strategy:

• Primary: Groq (generous free tier)
• Secondary: Google Gemini (free tier)
• Tertiary: GitHub Models (free for GitHub users)
• Quaternary: Ollama (local, free)

Best For:

• Open-source projects
• Learning and experimentation
• Personal projects
• Zero-budget deployments

Cost: $0/day (completely free)

---

Quick Start

1. Copy a Configuration

# For local development
cp examples/configs/local-first.yaml .aqe/providers.yaml

# For production
cp examples/configs/hosted-only.yaml .aqe/providers.yaml

# For free tier
cp examples/configs/free-tier.yaml .aqe/providers.yaml

2. Set Environment Variables

# Required for hosted providers
export GROQ_API_KEY="gsk_..."
export OPENROUTER_API_KEY="sk-or-v1-..."
export ANTHROPIC_API_KEY="sk-ant-..."
export GOOGLE_API_KEY="AIza..."
export GITHUB_TOKEN="ghp_..."

# Optional: Override deployment mode
export LLM_MODE="local_first"

# Optional: Explicit provider selection
export LLM_PROVIDER="groq"

3. Initialize Agentic QE

# The configuration is automatically loaded
aqe init

# Verify configuration
aqe config show

---

Configuration Syntax

Basic Structure

mode: local_first | hosted | free_only | hybrid

providers:
  - type: ollama | groq | openrouter | claude | google | github
    enabled: true
    priority: 10  # Lower = higher priority
    apiKey: ${ENV_VAR}  # Environment variable interpolation
    baseUrl: https://api.example.com
    defaultModel: model-name

    modelOverrides:
      test-generation: specialized-model
      code-review: another-model

    costPer1MTokens:
      input: 0.59
      output: 0.79

    limits:
      requestsPerMinute: 30
      tokensPerMinute: 20000

    fallbackProvider: backup-provider

fallbackChain:
  - primary-provider
  - secondary-provider
  - tertiary-provider

costBudget:
  daily: 20.0
  monthly: 500.0
  warnThreshold: 80
  enforceLimit: true

Environment Variable Interpolation

Use ${VAR_NAME} to inject environment variables:

providers:
  - type: groq
    apiKey: ${GROQ_API_KEY}
    baseUrl: ${GROQ_BASE_URL}  # Optional override

Task-Specific Model Routing

Route different task types to specialized models:

modelOverrides:
  test-generation: qwen2.5-coder:7b      # Code-focused model
  code-review: deepseek-coder:6.7b       # Review-optimized
  documentation: llama3.2:3b             # Lightweight for docs
  security-scanning: claude-3.5-sonnet   # Premium for security

Rate Limiting

Configure rate limits to stay within provider quotas:

limits:
  requestsPerMinute: 30
  requestsPerDay: 14400
  tokensPerMinute: 20000
  tokensPerDay: 1000000

Cost Budgets

Set spending limits with automatic enforcement:

costBudget:
  daily: 5.0           # $5/day hard limit
  monthly: 100.0       # $100/month hard limit
  warnThreshold: 80    # Warn at 80% of budget
  enforceLimit: true   # Block requests when budget exceeded

---

4. hybrid.yaml

Use Case: Adaptive workloads, intelligent cost optimization

Strategy:

• Primary: Groq (fast inference for complex tasks)
• Secondary: Ollama (local for simple tasks)
• Tertiary: OpenRouter (premium quality when needed)

Best For:

• Teams with varied workloads
• Dynamic cost optimization
• Performance tuning based on task complexity

Cost: ~$5-20/day (adaptive based on usage)

---

Deployment Mode Comparison

Feature	local_first	hosted	free_only	hybrid
Primary Provider	Ollama	Groq	Groq/Google	Groq/Ollama
Fallback	OpenRouter	OpenRouter/Claude	Google/GitHub	Multi-tier
Cost	$0-5/day	$10-50/day	$0/day	$5-20/day
Latency	Low (local)	Medium	Variable	Adaptive
Privacy	High	Medium	Low	Medium-High
Reliability	Medium	High	Medium	High
Setup Complexity	High	Low	Low	Medium
Best For	Dev/Privacy	Production	Open Source	Adaptive Teams

---

Troubleshooting

Configuration Not Loading

# Check default locations
ls -la .aqe/providers.yaml
ls -la aqe.config.yaml

# Enable debug logging
export LOG_LEVEL=debug
aqe init

Provider Initialization Fails

# Verify API keys
echo $GROQ_API_KEY
echo $OPENROUTER_API_KEY

# Test provider health
aqe providers health

# Check specific provider
aqe providers test groq

Environment Variables Not Interpolating

# Ensure variables are exported
export GROQ_API_KEY="gsk_..."

# Verify interpolation syntax
cat .aqe/providers.yaml | grep '${GROQ_API_KEY}'

---

Next Steps

1. Customize: Edit the configuration to match your needs
2. Test: Run aqe init to validate the configuration
3. Monitor: Use aqe providers status to check health
4. Optimize: Adjust models and priorities based on usage

For more information, see:

• Configuration Reference
• Provider Guide
• Cost Optimization