Skip to main content
Dari’s managed router gives your application one hosted model endpoint. For each request, it chooses a model and reasoning level using the request and model capabilities. Dari owns endpoint authentication, provider execution, routing state, retries, fallbacks, telemetry, billing, and deployment.
The response is OpenAI-compatible. dari_routing.selected_model identifies the selected model, and dari_routing.reasoning_effort identifies its reasoning level.

​
Default Router

Every organization starts with a default router at /v1 that uses Dari-managed provider credentials. An organization is the workspace that owns your routers and Routing API keys. The starter router routes across Claude Fable 5, GPT-5.6 Sol, GLM-5.2, and DeepSeek V4 Flash. The default router is an ordinary router. Edit it, or create another router and set it as the default to change which router /v1 uses. Send Chat Completions, Responses, or Anthropic Messages requests with model: "dari/routing". New routers enable speculative routing by default. Eligible follow-up requests begin on the previous compatible model while Dari selects and records the route for the next turn.

​
Other Routers

Create another router when you want a separate model set, custom models, different provider credentials, or different routing behavior and evals. Use its rtr_... endpoint directly, or set it as the default for /v1.
  1. Create a router.
  2. Choose models and provider keys.
  3. Use a Routing API key from the same organization as the router.

​
Self-Host The Framework

Use the Router Framework when you want to self-host the endpoint and own credentials, policies, routing state, and operational behavior. The framework and managed platform share routing concepts, but only the managed platform supplies hosted infrastructure and automatic speculative orchestration.