CLI Reference

This section documents all available CLI commands in the Juice CLI tool.

Core Commands

check

Command: check [options]

Description: Checks the installation for common issues

Parameters:

• --json: Format output as JSON

Details:

• Checks installation in the default install directory
• Verifies NVIDIA driver versions and compatibility
• Reports any failures that would prevent GPU sharing

login

Command: login [controller]

Description: Log in to a controller

Parameters:

• --insecure-skip-tls-verify, -k: Skip verification of TLS certificates
• --token, -t: Token to use for authentication (must be empty or start with "m2m")
• controller (optional): The controller to connect to. Can be:
  • A predefined environment name
  • An HTTP/HTTPS URL
  • If omitted, uses the first environment from vendor configuration

logout

Command: logout

Description: Log the current user out of the controller

Note: Cannot logout while having an active session - must release the session first using session release

Agent Management

agent list

Command: agent list (alias: agent ls)

Description: List all the agents accessible to the user

Parameters:

• --json: Output in JSON format
• --fields: Specify fields to display (default fields: state, hostname, pool.name, id, numgpus)
• --sort: Sort output (default sort column: 1)

agent info

Command: agent info <id>

Description: Gets the information of a specific agent

Parameters:

• --json: Output in JSON format
• --fields: Specify fields to display (default fields: state, hostname, pool.name, id, numgpus)
• --sort: Sort output (default sort column: 1)
• id (required): The ID of the agent to get information about

agent run

Command: agent run [pool_id]

Description: Run the GPU Sharing agent in the foreground, sharing your GPU to a specified pool (not available in WSL)

Parameters:

• pool_id (required): ID of the pool to share GPUs with
• --exit-on-stdin: Exit when stdin is closed
• --disable-gpu-metrics: Disable GPU metrics collection
• --gpu-metrics-interval-ms: GPU metrics collection interval in milliseconds
• --listen-ip: IP address to listen on
• --cache-size: Size of the cache
• --gpu-ids: Comma-separated list of GPU UUIDs to share
• --labels: Comma-separated list of key=value labels
• --taints: Comma-separated list of key=value taints
• --stun-server: Comma-separated list of STUN servers
• --relay-server: Relay server address
• --force-relay: Force using relay server
• --gateway-port: Gateway port number (1-65535)
• --gateway-ips: Comma-separated list of gateway IP addresses

agent service

Command: agent service

Description: Manage the agent service (not available in WSL). These commands manage the agent service on Windows or Linux (via systemd).

agent service install

Command: agent service install [pool-id]

Description: Install the GPU Sharing agent as a service (not available in WSL)

Parameters:

• pool-id (required): ID of the pool to share GPUs with
• All parameters from agent run command are supported
• --cache-size: Size of the cache

GPU Management

gpu list

Command: gpu list (alias: gpu ls)

Description: List all the GPUs accessible to the user

Parameters:

• --json: Output in JSON format
• --fields: Specify fields to display (default fields: hostname, name, uuid, agentVersion, vram-h)
• --sort: Sort output (default sort columns: 3,2,1)

Token Management

m2m list

Command: m2m list [id] (alias: m2m ls)

Description: Lists all M2M tokens

Parameters:

• --json: Output in JSON format
• --fields: Specify fields to display (default fields: description, token, id, expiresAt)
• --sort: Sort output (default sort column: 1)
• id (optional): ID of a specific token to list

m2m create

Command: m2m create

Description: Creates a Machine-2-Machine authentication token for use on servers. Once you have created a M2M token pass it as a flag to the login command to authenticate with your token instead of going through a web based authentication flow.

Parameters:

• --description, -d (required): Description for this token
• --expire-days, -e: The number of days the token will be valid for (default: 30)

m2m remove

Command: m2m remove <id>

Description: Removes a M2M token

Parameters:

• id (required): ID of the token to remove

Pool Management

pool list

Command: pool list (alias: pool ls)

Description: List all the pools accessible to the user

Parameters:

• --json: Output in JSON format
• --fields: Specify fields to display (default fields: organization.name, name, id, sessions, gpuCount)
• --sort: Sort output (default sort columns: 3,2,5,4,1)

Release Management

release list

Command: release list (alias: release ls)

Description: List releases

Parameters:

• --json: Output in JSON format
• --fields: Specify fields to display (default fields: version, notes)
• --sort: Sort output

release download

Command: release download <version> (alias: release dl)

Description: Download a specific release

Parameters:

• version (required): Version of the release to download. If only a partial version is provided (without build info), it will match the first release that starts with that version.

Application Management

run

Command: run [flags] <application> [<application args>]

Description: Starts an application using the GPU(s) available to the user

Parameters:

• application (required): The application to run
• --ephemeral: Always requests a new session for the application
• --on-queue-timeout: When a queue timeout happens, [fail, continue] (default: fail)
• --on-connection-error: When a connection error happens, [fail, continue] (default: fail)
• --disable-compression: Disables compression
• --disable-cache: Disables caching
• --relaxed-api: Enables a relaxed API conformance mode which allows for greater performance at expense of compatibility
• --listen-ip: IPv4 address to bind to when listening for connections
• --disable-uvm-kernel-synchronization: Disable synchronizing UVM memory before/after kernel launches

Note: The following options are mutually exclusive: session, query, gpu-ids, pool-ids, match-labels, tolerates, num-gpus, ephemeral

Session Management

session list

Command: session list (alias: session ls)

Description: List all the user's sessions

Parameters:

• --all: List all sessions (by default only shows active session)
• --json: Output in JSON format
• --fields: Specify fields to display (default fields: state, id, expiresAt, latencyMs, uploadMbps, downloadMbps)
• --sort: Sort output (default sort column: 1)

session release

Command: session release

Description: Releases the active session

session request

Command: session request

Description: Requests a new session

Parameters:

• --gpu-ids: Comma separated list of GPU IDs to allocate
• --pool-ids: Comma separated list of pool IDs from which to allocate the session resources
• --match-labels: Comma separated list of key=value pairs
• --tolerates: Comma separated list of key=value pairs
• --wait-for: Maximum number of seconds to wait for a GPU (default: 0)
• --no-version-check: Disables version checking when allocating the session requirements
• --num-gpus: Number of GPUs required (default: 1)
• --session-timeout: Time for the session to be automatically returned to the pool if not in use (default: -1)
• --watch-pid: Watch the specified process and stop the session when it exits (default: -1)
• --dump-config: Dumps the client configuration to the console in JSON format

Note: The following options are mutually exclusive: gpu-ids, query

User Management

user info

Command: user info [id]

Description: Retrieves user information

Parameters:

• --json: Output in JSON format
• --fields: Specify fields to display (default fields: name, email, id)
• --sort: Sort output (default sort column: 1)
• id (optional): ID of the user to get information about. If not provided, gets current user's information.