Skip to main content
The models.yaml file acts as the internal DNS for your gateway. It maps the model names requested by your clients to the specific upstream providers capable of fulfilling them.

​
File Structure

models:
  # The model string your client sends in their JSON request
  gpt-4o-mini:
    providers:
      - openai
      - azure-openai
    strategy: round_robin
    owned_by: open-next-router

  claude-haiku-4-5:
    providers:
      - anthropic
    strategy: round_robin

​
Directives

  • providers: An array of provider names (these exact names must match your .conf files in the providers directory).
  • strategy: The load balancing strategy relative to the configured providers. Currently, round_robin is supported by default.
  • owned_by: (Optional) Overrides the internal owner metadata reported when querying /v1/models.

​
Routing Flow and Priority

When a request arrives at ONR, the gateway must determine which provider’s DSL to execute. It selects the upstream based on this strict priority:
  1. Forced Override (Highest): If the client explicitly maps a provider via the x-onr-provider HTTP header.
  2. Model Registry Matching: Iterating through your models.yaml dictionary to find a matching model string.
  3. Round Robin Execution: If the chosen model lists multiple providers, ONR will load-balance the requests across the array uniformly per request.