Your users, on call.

What SQUAT is

SQUAT (short for SQuad UAT, pronounced like “squad”) builds a squad of tester personas, and where each one comes from is recorded rather than assumed. A squad member may be built from material a real person supplied about themselves, from documentary observation or recollection, from deliberately invented material, or from a combination — and SQUAT always tells you which. That provenance is the point: a tester whose origin you can trace is worth more than one you have to take on faith. Recruiting interviews or imports the source, preserves claims, motivations, behavioral patterns, tensions, and communication calibration, and compiles a provenance-excluding persona your AI can play. Then you run rounds: each squad member attempts realistic tasks against your live product, in character, and an independent evaluator scores what actually happened.

Every result lands in a ledger — a longitudinal record that answers the question a single test never can: is my product getting better or worse over time?

Three ideas shape everything else:

• Bring your own compute. Your AI client — Claude, ChatGPT/Codex, LM Studio, or another compatible MCP client — does the thinking. SQUAT hosts the workspace, the testing methodology, and your history. You never pay SQUAT for model time.
• Protected automatically. Connected clients send ordinary content through the authenticated service. SQUAT protects customer content before storage and returns it only to authorized workspace roles.
• Accurate measurement. A test has to be able to fail. The persona that plays your user never grades its own work. And when a result can't be trusted, it's flagged — never quietly deleted.

How this manual is organized

The Getting started section takes you from nothing to a working, connected squad. Working with your squad is the day-to-day manual: recruiting, running rounds, measuring, and reviewing. Data & administration covers migration, upkeep, and security. Reference is where to look when something's stuck or a term is unfamiliar.

How you'll actually use SQUAT

SQUAT has no app of its own to learn. You talk to your AI client in plain language, and it operates SQUAT for you:

• “Recruit a tester” — SQUAT helps you create or import a governed synthetic persona and adds it to your squad.
• “Run a UAT round” — your squad members attempt scenarios against your product.
• “How did the round go?” — read the results and the trend.
• “Run a steward sweep” — an expert review of your product against your own recorded intent.
• “Clean up my workspace” — a periodic hygiene and backup sweep.
• “Build me a playbook” — turn a pile of feedback, or a question you want answered, into a staged research program you approve before anything runs.

Throughout the manual you'll occasionally see tool names like round_start in monospace. You never call these yourself — they're the operations your AI performs on your behalf, named here so you can recognize them when your AI mentions them.

Conventions used in this manual

The SQUAT server address shown throughout the examples is https://api.squat.pro/mcp — the production endpoint every hosted workspace uses. If your organization runs its own SQUAT server, substitute the address your operator gives you.