Understand agent experience. From your terminal.
Gauge Agents helps developer-tool teams measure how coding agents discover, choose, and use their products. The official CLI brings that Agent Led Growth workflow to the terminal you and your agents already use.
Gauge Agents CLI
Your next evaluation starts here
npm install -g @withgauge/cli
gauge auth login
gauge instructionsOfficial npm package: @withgauge/cli · Node.js 22.12 or later
Define what success looks like
Create a realistic user task and measurable scoring criteria. Choose the repository, agent configurations, and tools that reflect how developers actually use your product.
See where agents get stuck
Launch evaluations, follow runs, and inspect results and run evidence. Find the docs, onboarding steps, or product interactions that keep an agent from completing the task.
Measure the next improvement
Use the evidence to author changes, then evaluate again. Manage Agent Preference measurements alongside your evals to understand which products agents choose.
Install once. Put your product to the test.
Install the official package
With Node.js 22.12 or later installed, use npm to install @withgauge/cli globally. The package provides the gauge executable.
npm install -g @withgauge/cliConnect your Gauge workspace
Installing the package does not sign you in. Authenticate to your existing Gauge account, then select your organization. New to Gauge Agents? Use gauge onboard for guided setup.
gauge auth login gauge orgs use <org>Give your agent the instructions
Read the command instructions bundled with your installed CLI. Your coding agent can use them to discover the supported workflows, while command help provides the available options.
gauge instructions gauge --helpRun an eval and inspect the evidence
Create your task and scoring criteria using the guide, choose an agent configuration, and launch the evaluation. Replace the placeholders below with your eval and run IDs. Runs use your workspace credits.
gauge models list gauge evals list gauge evals run <id> gauge runs watch <run-id>