A support agent that investigates, then asks before it acts.
AgentDesk runs a real Claude tool-use loop over a seeded support database. It reads customers, invoices, and policy to figure out what is going on, remembers durable facts, and proposes any customer-facing write for a human to approve. Every step of its reasoning is shown below.
Investigationready
Ask the agent to resolve a support issue
The agent investigates with read tools, then proposes any customer-facing action for you to approve. Try a sample task below or type your own.