Skip to main content
Quick reference for auxiliary CLI utilities.

​
hud version

hud version
Shows the HUD CLI version.

​
hud set

hud set HUD_API_KEY=... ANTHROPIC_API_KEY=...
Saves KEY=VALUE pairs to ~/.hud/.env for default use in HUD.

​
hud cancel

Cancel remote evaluation jobs submitted via hud eval --remote. 
# Cancel all tasks in a job
hud cancel <job_id>

# Cancel a specific trace within a job
hud cancel <job_id> --trace-id <trace_id>

# Cancel ALL your active jobs (panic button)
hud cancel --all
Options:
  • <job_id> - Job ID to cancel (positional argument)
  • --trace-id, -t <trace_id> - Cancel a specific trace within a job
  • --all, -a - Cancel all your active jobs
  • --yes, -y - Skip confirmation prompt
Requires HUD_API_KEY to be set.

​
hud models

List available models from the HUD inference gateway. 
hud models [--json]
Options:
  • --json - Output as JSON

​
hud push

Push a locally built Docker image to the HUD registry. This is a lower-level command — most users should use hud deploy which handles the full build + push flow. 
hud push [directory]
ArgumentDefaultDescription
directory.Environment directory with hud.lock.yaml
The image must be built first with hud build. Push reads the lock file to determine which image to push.
hud push is hidden from the main help output. Use hud deploy for the standard workflow.

​
hud convert

Convert external task/dataset formats to HUD format. 
hud convert <path> [--from <format>] [--output <path>]
Options:
  • <path> - Path to source tasks/dataset directory
  • --from - Source format (auto, webarena, browsergym, etc.)
  • --output, -o - Output path for converted tasks

​
hud scenario

Test scenarios against a running MCP server. Requires hud dev running in another terminal.

​
List scenarios

hud scenario list
hud scenario list --url http://localhost:9000/mcp

​
Run setup (get the prompt)

hud scenario setup my-env:echo --args '{"message": "hello"}'

​
Grade (submit answer + evaluate)

hud scenario grade my-env:echo --answer "hello"

​
Full run (setup + grade)

hud scenario run my-env:echo --args '{"message": "hello"}' --answer "hello"
# Prompt: Repeat this back exactly: hello
# {"reward": 1.0, "done": true}
Options:
  • --url, -u - Server URL (default: http://localhost:8080/mcp)
  • --args, -a - JSON string of scenario arguments
  • --answer, -A - Answer to submit for grading