Skip to main content
You can onboard through the full Gauge web product or the Gauge CLI. Both options let you create Agent Preference and Agent Experience measurements, configure their environments, launch runs, and inspect results.

Onboard with the web product

To evaluate product changes before merging them, follow Pull request evaluations. The GitHub App runs committed cases against your candidate package, executable, docs preview, Skills, or files and reports the outcome on the PR. Sign in at agents.withgauge.com and complete the guided onboarding flow. You can create or join an organization, seed starter measurements, configure runs, and review results without installing the CLI.

Read your first results

At the end of onboarding, Gauge starts a first scan of your product. The page shows each session live while it runs. When Gauge has enough evidence, click See top issues. You do not have to wait for every session to finish. The results show the top 3 issues and the number of sessions each one affected.
  • Issues that you can fix with your docs, Skills, or CLI come first. The top issue has an Optimize the top issue button. See How optimizations work.
  • An issue with a Product decision badge comes from how the product is built or priced. A docs change will not fix it, so it has no Optimize button.
Gauge also sends an email with your top issues when they are ready, so you can leave the page. One day later, Gauge sends a summary email with your Agent Experience and Agent Preference results.

Onboard with the CLI

Prerequisites

Before you begin, install Node.js 22.13 or later.
1

Install the CLI

Install gauge from npm.
Run gauge update later to get the latest version.
2

Create or join an organization

Run the guided onboarding flow. Gauge opens your browser to sign you in.
The command can create a workspace and seed starter Agent Preference and Agent Experience measurements.
3

Use the hosted backend

The CLI runs agents locally by default. Set your default backend to hosted Gauge:
To run agents in Docker on your machine instead, see the CLI guide.
4

Select an organization

List the organizations you can access, then save one as your default.
5

Inspect your setup

View the organization summary and its measurements.
Run gauge --help or gauge <command> --help whenever you need the current options for a command.

Launch and inspect one eval

After onboarding, check your billing balance and create a small task with an observable outcome:
Copy the eval ID returned by the command and replace <eval-id> below. This eval runs only when you ask it to. Start a run with the settings you just saved:
Review the run count and confirm when prompted. Copy a run ID from the list, then inspect it:
A useful first result includes the recorded session, file change, and judged criterion. Execution success and a passing criterion are separate: wait for judging before reading the eval’s pass rate. See Run troubleshooting if either is missing. Next, follow Create and inspect an eval, configure runs for your codebase, or test an improvement.

Next steps

Learn how Gauge works

Understand measurements, run settings, preference scenarios, and judged results.

Automate with the CLI

Use JSON output, environment variables, and declarative specs in CI.