Skip to main content
Match rules are used to selectively apply different configurations depending on what API the client is targeting, or whether the request is streaming.

​
Syntax

match api = "<api-name>" stream = true { ... }
match api = "<api-name>" stream = false { ... }
match api = "<api-name>" { ... }

​
Routing Logic

  • Feature Scope (v0.1): Currently tests against the api type and an optional stream boolean.
  • Top-to-Bottom Execution: The first match wins based on appearance order in the file. Always put your most specific rules first, and general catch-all rules at the bottom.
  • Strict Rejection: If a provider is selected (e.g. via load balancing) but no match is found inside its config file, the request is rejected with HTTP 400. ONR does not silently fallback; you must explicitly define matching rules.

​
Supported API Values

The api identifier is aligned with standard OpenAI-style endpoints (and specific extensions):
  • completions
  • chat.completions
  • responses
  • claude.messages
  • embeddings
  • images.generations
  • images.edits
  • audio.speech
  • audio.transcriptions
  • audio.translations
  • gemini.generateContent (Gemini native: POST /v1beta/models/{model}:generateContent)
  • gemini.streamGenerateContent (Gemini native: POST /v1beta/models/{model}:streamGenerateContent?alt=sse)