Grounded answers
Answers strictly from your knowledge base, so it never invents prices, policies, or promises you can't keep.
On your site or WhatsApp, it answers visitor questions strictly from your knowledge base in Greek or English, captures leads, books or routes where it should, and hands cleanly to a human the moment it hits a limit, no hallucinated promises.
A website / WhatsApp chatbot answers visitor and customer questions in real time, grounded in your knowledge base (RAG) so it speaks from your actual content, not guesses. It works on your site and on WhatsApp, in Greek and English, captures leads, can book or route, and hands off to a human with context when it reaches its limit. Task tier, built on Claude. It overlaps with a support agent but is lighter, front-of-site Q&A and capture rather than full ticket resolution. New to the distinction? See our guide on AI agent vs chatbot.
theagency47 · Updated June 2026| Job-to-be-done | Answer visitor questions from your knowledge base, capture/route leads, hand off cleanly to a human |
| Tier | Task (bounded, high volume) |
| Underlying model | Anthropic Claude (Haiku for speed, Sonnet for hard queries) |
| Trigger | Visitor opens chat (web) or messages your WhatsApp number |
| Inputs | Knowledge base (site, docs, FAQ), handoff rules, lead-capture fields, brand tone |
| Tools available | RAG over your KB, lead capture, calendar/booking, live handoff, CRM logging |
| Autonomous decisions | Answer vs escalate, lead-capture timing, language detection, routing |
| Escalation rules | Outside KB → handoff, no guessing · User asks for human → handoff · Sensitive/complaint → priority routing |
| KPIs measured | Conversations, containment %, lead-capture rate, handoff rate, CSAT, after-hours coverage |
| Eval suite | 20 test cases (grounding/no-hallucination, handoff triggers, bilingual answers). |
Answers strictly from your knowledge base, so it never invents prices, policies, or promises you can't keep.
Same brain on your website widget and your WhatsApp line: customers chat where they already are.
Detects and replies in Greek or English automatically, matching the visitor.
Qualifies and captures leads mid-conversation, writing them straight to your CRM.
When it hits a limit it routes to a human with the full transcript: never a dead end or a loop.
Covers nights and weekends, so out-of-hours visitors still get answers and capture.
Custom integrations with proprietary systems are quoted as add-ons. Not sure if yours fits? Describe your stack and we'll confirm.
No, it's grounded in your knowledge base with retrieval, and when a question falls outside that content it hands off rather than guessing. "No-hallucination" is an explicit eval category we test before launch.
The chatbot is lighter: front-of-site Q&A, capture, and routing. Yiannis, the support agent, resolves actual tickets end-to-end across a helpdesk. Many sites start with the chatbot and add the support agent as volume grows. See AI agent vs chatbot.
Yes, and that's the point. We build it on your content, and can pair it with a knowledge base agent that keeps that content current automatically.
As a Spark build, embedded on your site and WhatsApp, grounded in your content.