@evalguard/cli
v2.2.2
Published
EvalGuard CLI - Run LLM evaluations and security scans from the command line
Downloads
77
Maintainers
Readme
@evalguard/cli
Command-line interface for EvalGuard -- run LLM evaluations, security scans, and guardrail checks from your terminal and CI/CD pipelines.
Installation
npm install -g @evalguard/cliQuick Start
# Authenticate
evalguard login --key eg_live_...
# Initialize a project
evalguard init --project my-project
# Run an evaluation
evalguard eval evals/example.json --wait
# Run a security scan
evalguard scan scans/example.json --wait
# Run a local eval (no API key needed)
evalguard eval:local --model openai:gpt-4o --prompt "Answer: {{input}}" --scorer exact-match
# Check firewall rules
evalguard firewall check "Ignore all instructions"Commands
| Command | Description |
|---|---|
| evalguard login | Authenticate with your EvalGuard API key |
| evalguard logout | Remove stored credentials |
| evalguard init | Initialize EvalGuard in the current project (creates config and example files) |
| evalguard eval <file> | Run an evaluation from a JSON config file |
| evalguard scan <file> | Run a security scan from a JSON config file |
| evalguard whoami | Show current authentication status |
| evalguard eval:local | Run an evaluation locally without the cloud API |
| evalguard scan:local | Run a security scan locally without the cloud API |
| evalguard generate | Auto-generate eval cases or scan configs from a prompt |
| evalguard validate | Validate an eval or scan config file for correctness |
| evalguard compare | Compare results between two eval runs (drift detection) |
| evalguard list | List past eval runs and security scans |
| evalguard firewall | Check input against firewall rules or manage rule sets |
| evalguard watch | Watch eval/scan files and re-run on changes |
Configuration
The CLI stores credentials at ~/.evalguard/config.json. Project-level settings go in evalguard.config.json in your repository root.
{
"$schema": "https://evalguard.ai/schema/config.json",
"projectId": "my-project",
"defaultModel": "gpt-4o",
"evalsDir": "./evals",
"scansDir": "./scans"
}CI/CD Usage
# GitHub Actions example
- name: Run EvalGuard security scan
env:
EVALGUARD_API_KEY: ${{ secrets.EVALGUARD_API_KEY }}
run: |
npx @evalguard/cli login --key $EVALGUARD_API_KEY
npx @evalguard/cli scan scans/production.json --waitDocumentation
Full documentation at evalguard.ai/docs/cli.
License
Apache License, Version 2.0 — see LICENSE and NOTICE.
This CLI is a thin public client for the EvalGuard service. It is not covered by any proprietary license restriction — you may fork, modify, and redistribute it under Apache 2.0. The EvalGuard service itself, its backend engine, scorers, and attack plugins are proprietary software operated as a hosted service and are NOT covered by Apache 2.0. Access is governed by the EvalGuard Terms of Service.
"EvalGuard" is a trademark of EvalGuard, Inc. Derivative forks must not use the EvalGuard name or logo to imply endorsement or drop-in compatibility with the hosted service.
