Command-line reference#

This reference is generated from the installed commands in each top-level sample’s pyproject.toml under agent-samples/ and model-server-samples/, and the literal argparse declarations in its entry-point module. Worker and service commands are launcher-managed process protocols, so they are intentionally excluded from the user-facing catalog.

lab_instrument_monitoring#

Launch lab instrument monitoring with reused model dependencies.

Run from agent-samples/lab-instrument-monitoring/:

$ uv run lab_instrument_monitoring [--expose-web-events] [--capture]

Options

--expose-web-events: bind the unauthenticated event viewer to all IPv4 interfaces instead of loopback
--capture: record participant video, bidirectional audio, and data-channel traffic

simple_vlm_example#

simple-vlm-example orchestrator — vision Q&A over voice or text.

Run from agent-samples/simple-vlm-example/:

$ uv run simple_vlm_example [--capture]

Options

--capture: record participant video, bidirectional audio, and data-channel traffic

tea_making_sample#

Launch native tea guidance with reused model services.

Run from agent-samples/tea-making-sample/:

$ uv run tea_making_sample [--expose-web-events] [--capture]

Options

--expose-web-events: bind the unauthenticated event viewer to all IPv4 interfaces instead of loopback
--capture: record participant video, bidirectional audio, and data-channel traffic

xr_render_demo#

xr-render-demo orchestrator. Runs the process stack for this sample.

Run from agent-samples/xr-render-demo/:

$ uv run xr_render_demo [--capture]

Options

--capture: record participant video, bidirectional audio, and data-channel traffic

clef_flash_model#

Launch the persistent local Clef-Flash 9B SystemOne service.

Run from model-server-samples/clef-flash/:

$ uv run clef_flash_model [--stop]

Options

--stop: Stop only the Clef service configured by yaml/clef_server.yaml.

model_servers#

model-servers orchestrator — starts one shared AI inference stack and exits.

Run from model-server-samples/model-servers/:

$ uv run model_servers [--stop] [--models PATH] [--allow-anonymous] [--gpu-profile NAME]

Options

--stop: Stop every persisted model-server stack and exit.
--models PATH: Custom local deployment JSON; defaults to yaml/models.default.json.
--allow-anonymous: Start without HF_TOKEN (unauthenticated downloads of the multi-GB checkpoints may stall indefinitely).
--gpu-profile NAME: Use a named YAML GPU profile instead of automatic detection. Intended for explicitly reviewed custom hardware profiles.

model_servers_nim#

Start a persistent NIM model stack, with hardware-specific local fallbacks.

Run from model-server-samples/model-servers-nim/:

$ uv run model_servers_nim [--stop] [--dry-run] [--export-models PATH] [--gpu-profile NAME] [--models PATH] [--allow-anonymous]

Options

--stop: Stop this stack’s model ports.
--dry-run: Validate and print the launch plan without credentials, Docker, or model downloads.
--export-models PATH: Write a reusable client models JSON and exit without launching servers.
--gpu-profile NAME: Select a hardware profile; otherwise detect Blackwell, dual Ada, or Spark.
--models PATH: Custom deployment JSON; defaults to yaml/<gpu-profile>/models.json.
--allow-anonymous: Allow Hugging Face downloads without HF_TOKEN for local fallback services.