llama-swap
Browse the docs

Docs / Upstreams

audio.cpp UI with the /audiocpp endpoint

Name a model audiocpp_auto and use /audiocpp/ to reach audio.cpp's own UI and model management without naming a model in each request.

audiocpp_server has its own web UI and model management. /audiocpp/ is a passthrough to that server. It is wired to one reserved model ID, audiocpp_auto, so requests do not have to say which model to use.

Setup

Name the model exactly audiocpp_auto and point a browser at http://localhost:8080/audiocpp/.

models:
  audiocpp_auto:
    # audio.cpp is compiled into the unified image
    cmd: >
      docker run --rm --runtime nvidia
      -p ${PORT}:8080
      -v /mnt/nvme/audio.cpp:/models
      --entrypoint audiocpp_server
      ghcr.io/mostlygeek/llama-swap:unified-cuda-rootless
      --ui --host 0.0.0.0 --ui-management

/audiocpp (no trailing slash) redirects to /audiocpp/ and keeps the query string. API key checks apply as they do on other endpoints. The path after /audiocpp is sent to audio.cpp unchanged, so /audiocpp/v1/tasks/run reaches audio.cpp's /v1/tasks/run.

Unlike comfyui_auto, nothing else is forced on this model. It is an ordinary model that happens to have a fixed URL.

What goes wrong

  • 404 local model audiocpp_auto not found. The model is named something else, or it is defined on a peer. /audiocpp/ only serves a local model with that exact ID. Aliases do not work here — the lookup is by model ID.
  • The page loads but model management is missing. Start audiocpp_server with --ui --ui-management.
  • The page cannot be reached from the host. The server must listen on 0.0.0.0 inside the container, and the container port must match -p ${PORT}:8080.
  • The model never becomes ready. The default checkEndpoint is /health. Set checkEndpoint to a path the server answers, or none. See guides/model-runtime/troubleshooting-model-wont-load.
  • To use a different model ID or run more than one. Serve it through /upstream/<model>/ instead.
  • guides/connectivity/upstream-passthrough — the general passthrough endpoint
  • guides/upstreams/comfyui — the same pattern for ComfyUI