Summary
I demonstrate a Bookly customer‑service agent running locally, showing how it handles a user request to return books. The demo begins with the chat UI and quickly encounters a hard guardrail because the intent is unclear, prompting clarification. I then authenticate the session by entering a PIN, which flips the internal flag and unlocks the tool that accesses the SaaS backend. Once authenticated, the agent checks return eligibility, confirms I qualify, and processes the refund, including a loyalty‑based early‑refund check. A resolution metric triggers a proactive concierge pop‑up, illustrating a bug and the system’s ability to surface relevant prior‑call context. Finally, I show how the agent reads previous traces to verify inventory before offering a purchase, highlighting real read/write, authenticated, hard‑guardrailed operations in a live customer‑service flow.