Onboarding Guide
Before your first audit, you complete a 5-step setup that defines your agent's profile. This profile is what Airi reads to design relevant, non-generic challenges.
Step 1 — Agent type
Select the category that best describes what your agent does — Customer Support, Sales, Legal, Healthcare, E-commerce, and 14 others. This anchors the behavioral frame Airi uses when simulating user interactions.
Step 2 — Commercial context & purpose
Specify whether the agent is deployed commercially and whether it serves real paying clients. Also select your primary use case: own business, company work, community project, or exploration. This affects the risk profile used in scoring.
Step 3 — Identity
Give your agent a name and select its operating language. The name appears throughout the Playground and reports. The language determines the language Airi will use during the audit session — it should match the language your agent actually speaks.
The platform language (EN/ES/ZH) is separate from the agent language. You can use PlayClaw in Spanish while auditing an agent that operates in Mandarin.
Step 4 — Audience context
Describe who uses your agent: their technical level, what they typically ask for, what frustrates them, and what a successful interaction looks like from their perspective. The more specific and honest this description, the more realistic the challenges Airi will generate.
Good example
"Non-technical small business owners who need help with billing questions, plan upgrades, and basic onboarding. They're often frustrated by jargon and expect clear, direct answers. They rarely read long responses."
Too vague
"Users who need support."
Step 5 — Technical brief
This is the most important step for audit precision. Define three things:
What it handles (scope)
The topics and tasks your agent is supposed to cover. Airi will probe whether the agent stays within this scope or wanders into adjacent territory.
Hard limits
Things the agent must never do — regardless of how the user asks. Airi will attempt to get the agent to cross these in round 5. The more specific, the better the test.
What success looks like (goal)
How a good interaction should end. This anchors the evaluation: did the agent actually achieve what it's supposed to achieve?
You can use the ✨ Suggest buttons on each field to have Airi generate a draft based on your previous steps. These are editable — always review and adjust them to match your actual agent behavior.