The hangar CLI¶

hangar is one binary for the server and the client. Install it as Binary describes. Every command is in the CLI reference.

hangar login --url https://hangar.example.com     # device sign-in in your browser
hangar login --url https://hangar.example.com --token hgr_...
hangar whoami
hangar logout

The URL and username are saved in ~/.config/hangar/config.yaml; the Token goes in the OS keychain (macOS Keychain, Windows Credential Manager or the Secret Service on Linux) where one is available, otherwise in that file with mode 0600. In scripts and CI, set HANGAR_CLI_URL and HANGAR_CLI_TOKEN instead, or pass --url and --token. HANGAR_CLI_CONFIG points the CLI at another configuration file, and HANGAR_CLI_NO_KEYCHAIN=1 keeps the Token in that file instead of the OS keychain.

Find and inspect¶

hangar search llama --filter format=gguf --filter quant=Q4_K_M   # the Arsenal
hangar search qwen --upstream --filter gguf --sort downloads     # Discover on huggingface.co
hangar info bartowski/Llama-3.2-3B-Instruct-GGUF                 # files, quants, Dependencies, Arsenal state
hangar info datasets/lhoestq/demo1                               # Configs, Splits, rows and sizes

Pull and download¶

hangar pull bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M
hangar pull hf-internal-testing/tiny-random-bert --include '*.json' --include '*.safetensors'
hangar pull Qwen/Qwen3-0.6B --server-only
hangar pull datasets/lhoestq/demo1 --to ./demo1

If the files are not in the Arsenal, hangar pull asks the Instance to Pull them and shows the progress, then downloads them into the Hugging Face cache ($HF_HUB_CACHE, default ~/.cache/huggingface/hub), in the same layout Python libraries use, so they reuse the files and so does the CLI. Downloads resume after an interruption, retry by themselves after a dropped connection or a busy proxy, and are checked against their SHA-256 or git object ID.

Upload and import¶

hangar push ./my-finetune acme/my-finetune --create --private
hangar import --from-hf-cache

See Uploading and Importing an existing cache.

Local tools¶

eval "$(hangar env)"                                          # HF_ENDPOINT and HF_TOKEN for Python tools
hangar lmstudio install bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M
hangar ollama install bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M

Jobs¶

hangar jobs list
hangar jobs watch <job-id>
hangar jobs cancel <job-id>

Scripting¶

Every command takes --json and prints JSON on standard output; progress goes to standard error. hangar server and hangar worker print only logs, so for them --json selects JSON logs, which is also the default. Exit codes: 0 success, 1 error, 2 usage error, 3 not found, 4 authentication error.