NexaDesk

AI Chatbots

Deploy intelligent chatbots trained on your business content

NexaDesk chatbots use retrieval-augmented generation (RAG) to answer visitor questions using your own content. Unlike generic chatbots, they only reference information you have trained them on, reducing hallucination and keeping responses on-brand.

How It Works

  1. Visitor sends a message via the chat widget or a connected channel
  2. Query is embedded into a vector representation
  3. Relevant content chunks are retrieved from your knowledge base
  4. AI generates a response using the retrieved context
  5. Response is delivered to the visitor in real time

Chatbot Configuration

Each chatbot has its own settings accessible under Chatbots > [Your Bot] > Settings:

Choose between available models. GPT-4o provides the best balance of quality and speed for most use cases.

Controls how creative the responses are:

  • 0.1 - 0.3 — Factual and consistent (recommended for support)
  • 0.4 - 0.6 — Balanced
  • 0.7 - 1.0 — More creative and varied (suitable for sales)

Define the chatbot's personality and boundaries. This prompt is prepended to every conversation. Include instructions about tone, what topics to avoid, and when to escalate to a human.

Set how many previous messages to include in the AI context. More context means better conversational continuity but higher latency and cost. Default is 10 messages.

Agent Handoff

When the AI cannot help, it can escalate to a human agent:

  • Automatic — The bot detects frustration or repeated questions and offers to connect with a human
  • Manual — The visitor clicks a "Talk to a human" button
  • Keyword-based — Specific phrases trigger an immediate handoff

During handoff, the full conversation history is passed to the agent so the visitor does not need to repeat themselves.

Analytics

Track chatbot performance in Analytics > Chatbots:

  • Total conversations handled per day/week/month
  • Resolution rate — Percentage of conversations resolved without human intervention
  • Average response time
  • Most asked questions
  • Unanswered queries — Questions where the AI could not find relevant content

Use the unanswered queries list to identify gaps in your training data and improve your knowledge base.

PreviousEmbed WidgetNextCRM Pipeline