AI agent
It answers only when it knows. Then it steps aside.
It reads what you give it — articles, pages, files, already-resolved conversations — and answers showing where the answer came from. When a question falls outside the perimeter it does not improvise: it hands over with the context assembled.
Answers are generated with Anthropic’s Claude models. Nothing from your workspace is used to train third-party models.
- 01
Grounded in your content
It searches your content by meaning rather than by keyword, then reranks the best of it. Every reply carries a "Sources" button, so an agent can see exactly what it was built on.
RAG with citations - 02
It looks at the photos
Customers send a picture of the receipt, the damage, the error on screen. McFly actually reads it, on any channel, instead of replying "I can't see images".
Multimodal vision - 03
It collects what you need
Email, name, phone: a form rendered inside the chat on the widget and the app, free text parsed on WhatsApp. Before answering, before handing over, or both.
Data collection - 04
It knows when to stop
"WHEN these conditions THEN escalate" rules evaluated before the model, a brake that breaks automatic loops, and a hand-off that routes to the right team based on audience, channel and contact attributes.
Governed hand-off - 05
It learns from your team
When an agent corrects an answer and the correction is approved, it becomes knowledge. Good human replies are mined from conversations and proposed as new content.
Continuous improvement
Playground
Ask the real questions before the agent ever faces a customer: same answer, same sources, without touching a single conversation.
Let’s talk
Half an hour, and you’ll know.
No recorded deck. You tell us how you answer today, we tell you what would change and what it costs. If it does not make sense, we say so.
Mon–Fri, 09:00–13:00 CET