Getting Started

This guide walks you through installing OpenLatch, authenticating, and running your first protected AI agent session.

Step 1: Install

Run the one-line installer:

npx @openlatch/client init

npx is part of Node.js. On a machine without it, run the script for your platform instead — no runtime required.

# macOS and Linux
curl -LsSf https://openlatch.ai/install.sh | sh

# Windows
powershell -ExecutionPolicy Bypass -c "irm https://openlatch.ai/install.ps1 | iex"

Either way, the installer will:

1. Download the OpenLatch client
2. Detect your installed AI coding agents
3. Configure tool call hooks automatically

The scripts install for your user, with no administrator rights. See Installation for where the binaries land and what the PATH change means for your current shell.

Step 2: Authenticate

Link this machine to your OpenLatch account:

openlatch system auth login

This opens app.openlatch.ai in your browser. Sign in the way your organization signs in, and the browser hands the credential back to the client, which stores it in your operating system’s keychain. The command prints the organization the machine is now linked to.

Step 3: Start a Protected Session

Launch your AI coding agent as usual. OpenLatch hooks activate automatically:

# Example with Claude Code
claude

# OpenLatch is now monitoring tool calls in the background

Step 4: View Your Dashboard

Open the OpenLatch dashboard to see events from your session:

https://app.openlatch.ai

The dashboard shows:

• A timeline of intercepted tool calls
• Verdicts for each action (allowed, blocked, flagged)
• Details about each action the agent attempted

What Happens Next

OpenLatch runs silently in the background, capturing every action your agent takes into one behavioral-event stream. When enforcement returns a deny verdict, the agent receives a clear error message explaining why the action was stopped.

For a deeper understanding of the detection pipeline, see How It Works.