Skip to main content
The AI Gateway provides two flags to restrict which providers and models clients can use. This is useful for cost control, compliance, and ensuring applications only use approved AI services.

​
Restricting providers

​
only_allow_configured_providers

When set to true, only providers explicitly listed in your configuration are allowed. Requests to other providers are rejected. 
on_http_request:
  - type: ai-gateway
    config:
      only_allow_configured_providers: true
      providers:
        - id: "openai"
        - id: "anthropic"
With this configuration:
  • ✅ Requests to OpenAI models work
  • ✅ Requests to Anthropic models work
  • ❌ Requests to Google, DeepSeek, or other providers are rejected

​
Disabling specific providers

You can also disable individual providers while allowing others: 
providers:
  - id: "openai"
    disabled: false
  - id: "anthropic"
    disabled: true  # Temporarily disabled

​
Restricting models

​
only_allow_configured_models

When set to true, only models explicitly listed in your provider configurations are allowed. Requests for other models are rejected. 
on_http_request:
  - type: ai-gateway
    config:
      only_allow_configured_models: true
      providers:
        - id: "openai"
          models:
            - id: "gpt-4o"
            - id: "gpt-4o-mini"
With this configuration:
  • ✅ Requests for gpt-4o work
  • ✅ Requests for gpt-4o-mini work
  • ❌ Requests for gpt-3.5-turbo or other models are rejected

​
Disabling specific models

Disable individual models without removing them from configuration: 
providers:
  - id: "openai"
    models:
      - id: "gpt-4o"
        disabled: false
      - id: "gpt-4o-mini"
        disabled: false
      - id: "gpt-3.5-turbo"
        disabled: true  # Not allowed

​
Combined restrictions

For maximum control, combine both flags: 
on_http_request:
  - type: ai-gateway
    config:
      only_allow_configured_providers: true
      only_allow_configured_models: true
      
      providers:
        - id: "openai"
          api_keys:
            - value: ${secrets.get('openai', 'key')}
          models:
            - id: "gpt-4o"
            - id: "gpt-4o-mini"
        
        - id: "anthropic"
          api_keys:
            - value: ${secrets.get('anthropic', 'key')}
          models:
            - id: "claude-3-5-sonnet-20241022"
This configuration creates a strict allowlist:
  • Only OpenAI and Anthropic providers
  • Only three specific models across those providers
  • All other requests are rejected with clear error messages

​
Use cases

​
Cost control

Limit access to expensive models: 
only_allow_configured_models: true
providers:
  - id: "openai"
    models:
      - id: "gpt-4o-mini"  # Cost-effective
      # gpt-4o excluded - too expensive for general use

​
Compliance

Ensure only approved providers are used: 
only_allow_configured_providers: true
providers:
  - id: "openai"
    metadata:
      compliance: "soc2"
  # Other providers not approved by security team

​
Development versus production

Development allows all models: 
# dev.yaml
only_allow_configured_models: false
Production restricts to approved models: 
# prod.yaml
only_allow_configured_models: true
providers:
  - id: "openai"
    models:
      - id: "gpt-4o"
        metadata:
          approved_for_production: true

​
Team-specific access

Different gateway endpoints for different teams: 
# ml-team-gateway.yaml
only_allow_configured_providers: true
providers:
  - id: "openai"
  - id: "anthropic"
  - id: "google"

# support-team-gateway.yaml
only_allow_configured_providers: true
only_allow_configured_models: true
providers:
  - id: "openai"
    models:
      - id: "gpt-4o-mini"  # Only basic model for support chatbot

​
Error messages

When a request is rejected, the gateway returns a clear error: Provider not allowed: 
{
  "error": {
    "message": "Provider 'google' is not allowed by gateway configuration",
    "type": "provider_not_allowed"
  }
}
Model not allowed: 
{
  "error": {
    "message": "Model 'gpt-3.5-turbo' is not allowed by gateway configuration",
    "type": "model_not_allowed"
  }
}

​
Next steps