goethe
Serve local models and an API gateway from the shell
TLDR
SYNOPSIS
goethe command [options]
DESCRIPTION
goethe manages a local model server and the HTTP gateway in front of it. Model ids are Hugging Face repository names such as `zai-org/GLM-5.3-Flash`. Case does not matter, and the on-disk folder form with a double underscore between owner and name is accepted too. A mistyped id gets a suggestion. Commands that list things accept --json.Every value follows its flag. A bare value prints the corrected command instead of running. GPUs are named `gpu0`, `gpu1`, and so on, matching `goethe gpu`.Exit codes: 0 success, 1 error, 2 wrong usage, 3 a confirmation was required but there is no terminal (pass --yes).The installer published with the project is a shell script from meshapplied.com, not a distro package. `goethe update` and `goethe uninstall` manage that install. `goethe doctor` is the check after setup.
PARAMETERS
start
Start Goethe, then walk through installing a model, serving it, and configuring access. Without a terminal, print the equivalent commandsup [--no-boot]
Start Goethe and the API gateway, and enable start at boot. Models that were loaded last time come back. --no-boot starts once and does not enable bootstatus
Whether Goethe is running, the API address, and the names and ids of API keysdown [--keep-engines]
Stop Goethe and its models, and disable start at boot. --keep-engines leaves models serving while Goethe itself restartsservice enable | disable | status | restore --on | --off
Control start-at-boot, and whether loaded models return when Goethe startsrecommend [--json]
Catalog models that fit the GPUs Goethe may useinstall --model id-or-folder [--load] [--ctx n] [--gpu gpuN...]
Download a Hugging Face repository, or register a local folder. --load serves it after install. --yes loads beside other GPU programs without asking. --hf-token asks for a token for this download only and does not store itmodels [--installed | --available] [--json]
Installed models, then the catalog. list is an aliasload --model id
Load a model and wait until it answers. Extra engine options go after `--`unload --model id
Unload a modelps [--json]
Running models, their GPUs, and in-flight requestslogs [--model id] [-f] [-n n]
Model log, or --daemon / --gateway for Goethe's own logs. The default line count is 100configure api [--mode local|vpn|public]
Choose who can reach the API: this machine, a VPN address, or a public addressapi [--json]
API address, authentication, served models, keys, and a curl examplekey [--create name] [--revoke id]
List keys, create one, or revoke by id. --raw prints only the new secret. --rpm defaults to 120 (0 is unlimited)gpu [--allow gpuN...] [--deny gpuN...]
Show GPUs and whether new models may use themdoctor
Check GPU, driver, compiler, Python, gateway, ports, and keys--help
Help for the top level or for one command (`goethe` command `--help`)--version
Print the version
CAVEATS
install --load and load occupy GPU memory. --yes skips the question about other programs already on that GPU. remove deletes files Goethe downloaded. A folder you only registered is unregistered, not deleted.key --create prints a secret. --raw is for scripts. Do not log it. configure api --mode public exposes the gateway beyond localhost.A package named `goethe` in a distro archive is not this server. Do not install an unrelated package to get this binary.
HISTORY
Goethe is Mesh Applied's local model runner and API gateway. The command reference lives in the repository as `docs/cli.md`, next to the user guide. The source repository is meshapplied/goethe.
SEE ALSO
python3(1)
