Sustain Mode

Sustain Mode automatically routes your AI requests to minimize carbon emissions while maintaining your quality requirements.

How Sustain Mode Works

Sustain Mode analyzes each request and:

1. Assesses Carbon Impact: Evaluates real-time grid carbon intensity
2. Scores Model Efficiency: Considers each model's carbon footprint per token
3. Maintains Quality: Ensures responses meet your minimum quality threshold
4. Optimizes Routing: Selects the most sustainable option that meets requirements

Enabling Sustain Mode

Project-Level Configuration

Enable Sustain Mode for all requests in a project:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.gateflow.ai/v1",
    api_key="gw_prod_your_key_here"
)

# Enable Sustain Mode in project settings (via Dashboard or API)
# Then all requests automatically use carbon-optimized routing
response = client.chat.completions.create(
    model="auto",  # Let GateFlow choose the most sustainable model
    messages=[{"role": "user", "content": "Analyze this sustainably"}]
)

Request-Level Configuration

Override project settings for specific requests:

# Enable Sustain Mode for this specific request
response = client.chat.completions.create(
    model="gpt-5.2",
    routing_mode="sustain_optimized",
    minimum_quality_score=85,  # Minimum acceptable quality (0-100)
    max_carbon_per_request_gco2=50,  # Optional: Set maximum carbon budget
    messages=[{"role": "user", "content": "Generate eco-friendly analysis"}]
)

Configuration Options

Quality Threshold

Set the minimum quality score (0-100) you're willing to accept for carbon optimization:

# More aggressive optimization (lower quality threshold)
response = client.chat.completions.create(
    model="auto",
    routing_mode="sustain_optimized",
    minimum_quality_score=80,  # Allow more carbon-efficient models
    messages=[{"role": "user", "content": "Draft content"}]
)

# Conservative optimization (higher quality threshold)
response = client.chat.completions.create(
    model="auto",
    routing_mode="sustain_optimized",
    minimum_quality_score=95,  # Prioritize quality over carbon savings
    messages=[{"role": "user", "content": "Critical business analysis"}]
)

Carbon Budget

Set maximum carbon emissions per request:

# Set carbon budget for the request
response = client.chat.completions.create(
    model="auto",
    routing_mode="sustain_optimized",
    max_carbon_per_request_gco2=25,  # 25 grams CO₂ equivalent max
    messages=[{"role": "user", "content": "Quick summary"}]
)

Fallback Behavior

Configure what happens when no sustainable alternative meets your requirements:

# Fallback options:
# - "use_requested_model": Use the originally requested model
# - "use_most_efficient": Use the most efficient model regardless of quality
# - "fail": Return error if no suitable model found

response = client.chat.completions.create(
    model="gpt-5.2",
    routing_mode="sustain_optimized",
    minimum_quality_score=90,
    fallback_behavior="use_requested_model",  # Default: fall back to requested model
    messages=[{"role": "user", "content": "Important analysis"}]
)

Sustain Mode Response Data

Sustain Mode enhances responses with carbon tracking data:

response = client.chat.completions.create(
    model="auto",
    routing_mode="sustain_optimized",
    messages=[{"role": "user", "content": "Analyze sustainability trends"}]
)

# Access sustainability metrics
print(f"Model used: {response.model}")
print(f"Original model requested: {response.sustainability.original_model}")
print(f"Carbon footprint: {response.sustainability.carbon_gco2e} gCO₂e")
print(f"Carbon saved: {response.sustainability.carbon_saved_gco2e} gCO₂e")
print(f"Quality score: {response.sustainability.quality_score}")
print(f"Grid carbon intensity: {response.sustainability.grid_carbon_intensity} gCO₂/kWh")
print(f"Sustain mode active: {response.sustainability.sustain_mode_active}")

Best Practices

When to Use Sustain Mode

✅ Ideal for:

• Non-critical analysis and content generation
• Internal reports and summaries
• Draft content creation
• Background processing tasks
• Batch operations

❌ Use with caution for:

• Customer-facing critical responses
• High-stakes decision making
• Legal or compliance-related content
• Mission-critical applications

Gradual Adoption Strategy

1. Start with monitoring: Enable Sustain Mode in "monitor-only" mode to see potential savings
2. Test with non-critical workloads: Try on internal tools and background processes
3. Adjust quality thresholds: Find the right balance for your use cases
4. Monitor results: Track carbon savings and quality impact in the dashboard
5. Expand usage: Gradually apply to more workloads as you gain confidence

Advanced Configuration

Task-Type Exclusions

Exclude specific task types from Sustain Mode:

# Configure in project settings (via Dashboard or API)
excluded_task_types = [
    "legal_analysis",
    "financial_advice",
    "medical_diagnosis",
    "customer_support_critical"
]

Region-Specific Policies

Optimize for specific geographic regions:

# Prioritize providers with data centers in low-carbon regions
region_policy = {
    "preferred_regions": ["nordics", "canada", "france"],
    "avoid_regions": ["coal-heavy-region"]
}

Troubleshooting

"No suitable sustainable model found"

Causes:

• Quality threshold set too high
• Carbon budget too restrictive
• Requested model has no sustainable alternatives
• All providers experiencing high carbon intensity

Solutions:

1. Lower the minimum_quality_score
2. Increase the max_carbon_per_request_gco2
3. Adjust fallback_behavior to "use_requested_model"
4. Add more providers to increase options
5. Check grid carbon intensity in your region

Performance Impact

Sustain Mode adds minimal overhead:

• Routing decision: ~50-100ms additional latency
• Carbon data collection: ~20-50ms
• Total impact: Typically <150ms per request

For latency-sensitive applications, consider:

• Using routing_mode="performance_optimized" for critical paths
• Enabling Sustain Mode only for background processes
• Using asynchronous requests with time-shifted execution

Sustain Mode vs Manual Optimization

Feature	Sustain Mode	Manual Optimization
Carbon Reduction	Automatic, real-time	Requires manual research
Quality Maintenance	Guaranteed thresholds	Manual quality checking
Provider Selection	Automatic, multi-provider	Manual provider choice
Grid Awareness	Real-time carbon intensity	Static assumptions
Reporting	Automatic tracking	Manual tracking required
Maintenance	Automatic updates	Manual updates needed

Next Steps

• View Sustainability Dashboard - Monitor your carbon savings
• Configure Time-Shifted Execution - Defer requests to low-carbon periods
• Explore Provider Efficiency - Understand model carbon profiles