Skip to main content
Use this page as the customer-facing reference for harmstack flags.
This page intentionally shows canonical kebab-case flags only (for example --job-id, --benchmark-id, --unit-count).

Global and shared flags

These flags apply to the harmstack root command and harmstack init.

Authentication and endpoint targeting

string
Your Harmstack API key for account and job APIs.Env fallback: HARMSTACK_API_KEYExample:
string
URL of the model API endpoint to benchmark.Example:
string
Bearer token for your target model endpoint. Required when using --consentandskip.Env fallback: TARGET_MODEL_API_KEYExample:
string
default:"openai"
API shape for your target endpoint.Accepted values:
  • openai
  • openai_responses
  • gemini
  • raw
Example:
string
Model name used in requests. Ignored when --provider=raw.Example:

Benchmark selection and run behavior

integer[]
Benchmark IDs to run. Repeat the flag or pass a comma-separated list.Examples:
integer[]
Number of human-annotated unit tests per benchmark job (1 to 10). Must align with --benchmark-id order and length.Example:
boolean
default:"false"
Skip interactive prompts and run non-interactively. Recommended for CI and scripting.Example:
string[]
Optional HTTP headers added to every request to your model endpoint. Repeat as needed.Example:
boolean
default:"false"
Run the Haystack benchmarking flow directly from the root harmstack command.Example:

harmstack compare-jobs flags

string
UUID of the first job.Example:
string
UUID of the second job.Example:

harmstack list-jobs flags

string
default:"table"
Output format.Accepted values: table, csvExample:
integer
default:"10"
Maximum jobs to return.Example:
string
default:"completed"
Status filter.Accepted values: completed, failed, allExample:

harmstack show-job flags

string
UUID of the job to inspect.Examples:

harmstack stats flags

integer
default:"30"
Number of recent completed jobs to include in aggregate calculations.Example:
string
Date filter in YYYY-MM-DD format.Example: