Skip to main content
Aria Compute ROUTER is a self-hosted OpenAI-compatible inference gateway. It runs three parallel decisioners: a semantic router that matches requests against a YAML v0.3 recipe using routing signals; a lightweight builtin agent router that runs an in-process tool loop with fixed tools and limited turns; and an afm-d router that calls aria-engine System One Choice (AFM_ENGINE_URL). Gold entrypoints are ariacompute/semantic-auto, ariacompute/agent-auto, and ariacompute/afm-d-auto. One request never runs more than one decisioner. The router exposes a data plane for inference clients and a management plane for operators. The data plane implements the standard OpenAI /v1/models and /v1/chat/completions endpoints, so existing OpenAI SDKs work unchanged. The management plane covers authentication, routing configuration, cost tracking, and key management under /v1/router/*. The router evaluates shared provider models and hard constraints (cost, latency, capability) at request time. You configure entrypoints, providers, and routing rules in a YAML v0.3 config file that the router reloads dynamically.

Authentication

How to authenticate with the router: Bearer tokens for data plane, sessions or API keys for management.

Chat Completions

Send chat requests through the router with standard OpenAI parameters and streaming support.

List Models

Discover available entrypoints and provider models via the OpenAI-compatible models endpoint.

Auth Endpoints

Register, log in, manage users, and configure OAuth for the router dashboard.

Base URL

The router is self-hosted. Replace <router-host> with the host where your router runs.