Skip to main content
Open Next Router is fundamentally designed to provide an OpenAI-compatible surface area for any downstream client, while managing the complexity of routing those requests to arbitrary providers behind the scenes.

​
Core API Contract

Clients interacting with ONR should generally assume they are talking to a standard OpenAI completion, chat, or embedding endpoint. ONR sits in the middle and parses the standard shape:
  • POST /v1/chat/completions
  • POST /v1/embeddings
  • GET /v1/models

​
Dynamic Provider Override

Normally, ONR picks which upstream provider to route to by looking at the model parameter inside the JSON body and matching it against your models.yaml registry. However, a client can explicitly force a request to hit a specific upstream provider using the custom override header: 
x-onr-provider: anthropic
If present, this header bypasses models.yaml entirely and executes the DSL defined in config/providers/anthropic.conf.

​
Streaming Guarantee

Streaming LLM responses is notoriously tricky due to varying Server-Sent Event (SSE) framings, chunk layouts, and closing signals across different LLM providers. ONR guarantees that if a client sends a standard OpenAI stream: true flag in the JSON body:
  1. It will establish a persistent text/event-stream with the client.
  2. It will apply the respective sse_parse transformations defined in the provider’s DSL.
  3. It will emit cleanly formatted, OpenAI-schema-compliant delta chunks back to the client.