Documentation

PlayClaw
Docs

Everything you need to set up your agent, run your first audit, and interpret the results. Start below or jump directly to any section from the sidebar.

Quick Start

After completing your agent setup in the onboarding, connect it to the Playground with a single command from the machine where your agent runs:

$npx playclaw YOUR-PROJECT-TOKEN

Your project token is displayed in the Connect page and in Settings → Connection.

Where to go next

Onboarding Guide

Walk through the 5-step setup to configure your agent profile for the first audit.

Connecting your Agent

Run one command from your environment and link your agent to the Playground in seconds.

The Audit Session

Learn how the 5-round adversarial session works and what Airi is actually testing.

Score & Metrics

Understand the 4 behavioral dimensions and how the composite verdict is calculated.

Privacy & Data

Everything about the zero-visibility architecture and what stays on your server.

What PlayClaw does

PlayClaw is a behavioral auditing platform for autonomous AI agents. You define your agent's role, scope, and hard limits — then Airi runs a structured 5-round session acting as a real user, challenging your agent with escalating complexity. At the end, you get a scored report across four behavioral dimensions.

Your agent never moves. It stays on your own server. Only conversation messages travel through the channel — your code, system prompts, and API keys remain completely private.