The right model
for every prompt.

vllm-sr/auto reads each request and sends it to the open model and effort it needs, behind one OpenAI-compatible API.

Watch it choose.

Pick an example. You see the exact request, the router's real signal results, the lane, the model and its effort, then the answer as it streams.

1 · Request

Pick an example to see the question.

Your first run creates a free API key. By continuing, you accept the Terms and Privacy Policy.

2 · Signals

3 · Decision

    4 · Model and effort
    Mission

    Intelligence, routed.

    Every request deserves the right model: not always the biggest, never a guess. We build that decision in the open, for open models, on hardware anyone can run.

    vLLM Semantic Router on GitHub Read the docs

    1. Decide fast

      One typed call answers every routing question before the first token.

    2. Reason where it pays

      Thinking is a cost. Spend it on the proof, not on the fun fact.

    3. Open by default

      Open source, open-weight models and the open ROCm stack on AMD Instinct.

    Inside the routerHow it works

    How it works.

    System One understands the request. The router chooses the System Two model and reasoning effort.

    Client

    Your app or agent

    Any OpenAI SDK, model="vllm-sr/auto"

    vLLM Semantic Router
    Frontend

    One HTTPS endpoint

    • Listeners
    • Auth and keys
    • Chat Completions
    • System One typed API
    Signal and decision engine
    1. Signals
    2. Projections
    3. Decisions
    4. Algorithms
    5. Plugins

    vLLM Semantic Router
    Open source · Apache 2.0

    System One · decision model

    Hugging FaceVela-2.0-4B

    Reads intent, difficulty and reasoning needs in one pass.

    System Two · model pool
    AMD Instinct MI325X · powering fast decisions and deep reasoning
    System OneVela reads intent, difficulty and reasoning needs.
    System TwoOpen models answer with the right depth of reasoning.
    One APIThe router connects both through vllm-sr/auto.
    Event PyTorch

    PyTorch Conference 2026

    The talkFast Decisions, Deep Reasoning: Cost-Efficient Mixture-of-Models on AMD GPUs with vLLM Semantic Router

    Sponsor AMD

    The model pool and the decision model behind vllm-sr/auto run on AMD Instinct™ MI325X GPUs.

    AMD Instinct accelerators