scorable-cli 0.1.4

CLI for the Scorable API

Measurement & Control for LLM Automations

The scorable CLI is a powerful command-line tool for interacting with the Scorable API. It provides a convenient way to manage and execute your Judges directly from the terminal.

Installation

You can install the scorable CLI using the following command, which downloads and installs the script to /usr/local/bin:

curl -sSL https://scorable.ai/cli/install.sh | sh

Alternatively, you can install and run the CLI using uvx:

uvx scorable-cli judge list

Authentication

Before using the CLI, you must set your Scorable API key as an environment variable:

# Sign up for a free account at https://scorable.ai/register
export SCORABLE_API_KEY="your-api-key"

Temporary API keys

If no API key is set, the CLI can create a temporary key interactively and save it to ~/.scorable/settings.json as temporary_api_key. Permanent keys should be set via the SCORABLE_API_KEY environment variable, which takes precedence.

Usage

The CLI is organized into a main command, scorable, with subcommands for different functionalities. The primary resource you'll interact with is the judge.

Judge Management

All Judge-related commands are available under the scorable judge subcommand.

list

List all available Judges, with options for filtering and pagination.

scorable judge list

Options:

• --page-size: Number of results to return per page.
• --cursor: The pagination cursor value.
• --search: A search term to filter by.
• --name: Filter by exact judge name.
• --ordering: Which field to use for ordering the results.
• --is-preset / --not-is-preset: Filter by preset status.
• --is-public / --not-is-public: Filter by public status.
• --show-global / --not-show-global: Filter by global status.

get

Retrieve a specific Judge by its ID.

scorable judge get <judge_id>

create

Create a new Judge.

scorable judge create --name "My New Judge" --intent "To evaluate the quality of LLM responses."

Options:

• --name: The name for the new judge (required).
• --intent: The intent for the new judge (required).
• --stage: The stage for the new judge.
• --evaluator-references: JSON string for evaluator references. E.g., '[{"id": "eval-id"}]'

update

Update an existing Judge.

scorable judge update <judge_id> --name "My Updated Judge Name"

Options:

• --name: The new name for the judge.
• --stage: The new stage for the judge.
• --evaluator-references: JSON string to update evaluator references. Use "[]" to clear.

delete

Delete a Judge by its ID. You will be prompted for confirmation.

scorable judge delete <judge_id>

duplicate

Duplicate an existing Judge.

scorable judge duplicate <judge_id>

Judge Execution

execute

Execute a Judge with specific inputs.

scorable judge execute <judge_id> --request "What is the capital of France?" --response "Paris"

Options:

• --request: Request text.
• --response: Response text to evaluate.
• --contexts: JSON list of context strings. E.g., '["ctx1"]'
• --functions: JSON array for the "functions" field.
• --expected-output: Expected output text.
• --tag: Add one or more tags.

Using stdin input:

You can pipe input directly to the --response parameter:

echo "Paris" | scorable judge execute <judge_id> --request "What is the capital of France?"

cat response.txt | scorable judge execute <judge_id>

execute-by-name

Execute a Judge by its name.

scorable judge execute-by-name "My New Judge" --request "What is the capital of France?" --response "Paris"

Input can also be piped in similar way as with execute.

Prompt testing

Initialize a prompt testing experiment config and run it.

scorable prompt-test init
scorable prompt-test run

Development

This project uses uv for dependency management. To set up the development environment, run:

. .venv/bin/activate
uv pip sync pyproject.toml