Commands

Run lllm2 COMMAND --help for all options.

Command	Purpose
lllm2 --version	Print the installed package version and exit.
lllm2 service install	Install, enable and start the panel user service; accepts --host, --port and --no-start.
lllm2 or lllm2 panel	Start the panel; --host and --port control its listener.
lllm2 models [--json]	List installed GGUF checkpoints.
lllm2 engines list [--json]	List discovered llama-server builds.
lllm2 engines install cuda	Download this release’s pinned CUDA engine; accepts --name and --force (try CUDA 12 when driver checks fail).
lllm2 launch	Serve in the foreground using the selected model/backend’s saved workbench settings when available; accepts --model, --engine, --backend CUDA|Vulkan|modal, --device and --timeout.
lllm2 launch --backend modal --gpu TYPE	Serve on a Modal GPU through the local engine port; --model also accepts a catalogue id or name. --idle-timeout MINUTES accepts 0 to 1440 or off, and 0 or off disables it. Without it, lllm2 uses LLLM2_IDLE_TIMEOUT_MINUTES, then saved settings, then 30. See Modal GPU types.
lllm2 modal setup	Check Modal credentials and deploy the lllm2 app unless this version is already deployed.
lllm2 modal probe --gpu TYPE [--json]	Run a short, billed probe container, print the GPU name, memory, engine devices, the engine build (llama.cpp release, CUDA track and compiler) and its sha256, and save the result for later launches.
lllm2 modal download ID	Download a catalogue model and its companion files into the Modal Volume, with progress; a stored model downloads nothing. Ctrl-C cancels, and a rerun resumes.
lllm2 modal list [--json]	List running lllm2 serve calls with owner, elapsed time and estimated cost.
lllm2 modal stop [CALL_ID|--all] [--force]	Stop one call, or every orphaned call with --all; a call whose owner still heartbeats is skipped, and --force stops one such call by id.
lllm2 modal models [--json]	List models stored in the Modal Volume, with sizes and catalogue ids.
lllm2 modal remove ID|NAME	Delete a stored model, by catalogue id or the name models prints, with its companion files, unless a running call serves it.
lllm2 claude, lllm2 codex	Connect an installed coding-agent CLI to the running model.

Engine installation shows download progress, transfer speed, and estimated time remaining, followed by checksum, extraction, and startup status. Progress goes to stderr; stdout contains the installed engine path. Redirected output uses periodic text updates instead of an animated bar. If the server does not provide a download size, the display shows bytes transferred without a percentage or time estimate.