Milo gives your AI email agent fake customers with real email addresses. Six scripted personas (angry refunds, chargeback threats, silence, GDPR requests) reply over real SMTP, deterministically: same inputs, same replies, no LLM in the reply path. Plus-address run IDs and a JSON lookup API make every conversation assertable in CI. There is no signup: emailing refund-angry@joinmilo.email is the onboarding. Strict mode rebuffs vague promises. MCP server, llms.txt and an agent skill included.