Chapter 01
Intent understanding, not keyword matching.
A customer types 'do you have rooms for 2 nights next week with a balcony' — not 'rooms'. Most BSP bots match the word 'rooms' and reply with a generic rate list. Waaru's agent understands the underlying intent (availability + duration + preference) and answers correctly the first time.
Chapter 02
Grounded in your business, not the public internet.
The agent answers from your Company Brain — your website, product catalog, FAQs, policies — never invents pricing or policy. If a question falls outside the indexed content, the agent says so plainly and escalates instead of hallucinating.
Chapter 03
Multi-turn memory.
The agent remembers what was said ten messages back. 'What about the smaller room?' after asking about a suite resolves correctly to the smaller-of-two-rooms context. Memory persists per-contact across sessions too — preferences set in March still apply in May.
Chapter 04
Action-taking via typed tools.
When the conversation needs an action — order lookup, slot booking, payment link, refund — the agent calls the matching tool with arguments validated against a JSON schema. Every call is logged; every argument is checked before any external system is touched.
Chapter 05
Knows when to escalate.
When confidence drops below threshold, the agent escalates to the team inbox with the full transcript, an AI summary, the contact record, and any prior notes. The human picks up at exactly the right point. The customer never repeats themselves.