2024 to present
Menni.ai · Chatbot AI · Spjallmenni

Answering 100% of website chat, and completing the booking inside it

Spjallmenni greets every website visitor, understands the question in whatever language they typed it in, answers from the client's own knowledge base, and when the visitor wants to book, it makes the booking right there in the chat.

See it live at menni.ai
Menni.ai Chatbot AI
100%Of website chat handled by AI
100+Languages, auto-detected
14 daysFrom kickoff to live
24/7Always-on coverage
Live sitemenni.ai
IndustryB2B AI Customer Support
Engagement2024 to present
Scope of workAI EngineeringVoice AIPlatform ArchitectureMulti-tenant SaaSInfrastructure MigrationDashboard Product Build

Challenges and solutions

  1. Answering from real content instead of guessing

    The challenge

    Visitors ask very specific questions, like whether the lava show has space on April 24th for two adults. A generic bot either invents an answer or punts to a human, and follow-ups like "and who can I bring?" are meaningless read on their own. Either failure sends the customer back to waiting.

    What we shipped

    Replies are grounded in the client's own knowledge base, so they reflect real policies and real content. The agent tracks the whole conversation rather than the last message, so a follow-up resolves against what was actually being discussed.

  2. Making a support channel actually transact

    The challenge

    An agent that can answer but cannot act just relocates the work. The visitor still has to wait for a human to check availability, take the booking, or push a date back by a day.

    What we shipped

    The agent works directly with the Bókun booking platform in real time. Checking availability, booking, modifying and cancelling all happen inside the conversation, which turns a support channel into a booking channel.

  3. Getting live on any client's website in an afternoon

    The challenge

    Every client site is a different stack. A bespoke integration per customer would have capped how fast Menni could sign and onboard new business.

    What we shipped

    We shipped the widget two ways: an npm package for React, Next.js and Vue apps, and a one-line embed script for everything else. With Menni handling setup and knowledge ingestion, a client's typical path from kickoff to a live, trained agent is about 14 days.

  4. Staying fast and cheap as conversations get long

    The challenge

    Long chats balloon in both latency and cost, and every client on the platform needs their own knowledge, branding, tools and keys kept strictly apart from everyone else's.

    What we shipped

    The agent is multi-tenant by design, with isolated knowledge and configuration behind a single widget. Long conversations stay coherent without letting cost run away, and answers stream back token by token so the chat always feels immediate.

Before and after

  • BeforeVisitors waiting hours for a human replyAfterAnswers in seconds, at any hour
  • BeforeSupport available in office hours, in one languageAfter100+ languages, auto-detected, 24/7
  • BeforeBot answers, then a human completes the bookingAfterBooking completed inside the conversation
  • BeforeChat as a pure support costAfterChat as a channel that books revenue
How it works

What happens in a conversation

  1. 1
    Understand the whole conversation

    Follow-up questions are meaningless on their own. The agent keeps track of the full exchange, so it answers the real question rather than a fragment of one.

  2. 2
    Answer from real content

    Replies come from the client's own knowledge base, so they are grounded in the company's actual policies and content instead of invented on the spot.

  3. 3
    Take action through Bókun

    When the visitor wants to do something, check availability, book, modify or cancel, the agent works with the Bókun booking platform in real time and completes the task inside the chat instead of handing off.

  4. 4
    Learn from every chat

    After each conversation the system categorises it, detects the language, and flags spam or knowledge gaps, feeding the analytics and the knowledge-base improvement loop in the client dashboard.

The outcome

Chatbot AI handles 100% of website chat, answering in seconds instead of hours, in over 100 languages, without waking a human. It is the agent most Menni customers start with, the entry point to the wider platform, and the foundation the Phone and Email agents build on.

Technology stack

Install surface
@menni/chatbot (npm)One-line embed scriptReact / Next.js / Vue
AI
Vercel AI SDKOpenAIVector database
Backend & data
Nest.jsSupabase (Postgres)
Integrations
Bókun booking API

Part of the Menni.ai platform

Chatbot AI is one of three AI agents we built for Menni.ai.

Back to the full Menni.ai story

Want a support agent that closes the loop?

Answering questions is table stakes. We build agents that complete the task in the conversation, grounded in your own content. Tell us what yours would need to do.