Conversational Platform

AI agents that act instead of just answering

LoyJoy AI agents understand the request, use your approved knowledge and work with tools to complete the case in your systems. By chat and phone, with handover to your team when it matters.

How an AI agent works

A chatbot answers questions. An AI agent gets the request done, in four steps and within clear guardrails.

  1. 1

    Understand the request

    The agent recognises what the customer wants, even when phrased casually or on the phone.

    Example“I am moving on December 1. What do I need to do?”

  2. 2

    Use knowledge

    It searches your approved knowledge base and answers on that basis only.

    ExampleFinds the moving terms and the required details.

  3. 3

    Run the process

    It asks for missing data and uses its tools to execute the case in your systems.

    ExampleCreates the address change in the billing system.

  4. 4

    Resolve or hand over

    It confirms the result. Edge cases go to your team with full context.

    ExampleConfirmation by email, edge case goes to the service team.

Traceable in detail: every knowledge search and every tool call is visible in the conversation.

Tools make the difference

A language model on its own can only write. Tools let the agent search, work with real data, start processes and resolve requests. LoyJoy ships the key tools ready to use and connects your own systems via MCP.

Internal tools

Built into LoyJoy, switched on per agent without development.

Find knowledge

Answers with sources instead of guesses.

  • Knowledge search
  • Azure AI Search
  • Web search
  • Web fetch
  • Search and read files
  • Read customer uploads

Advise and sell

Matching products instead of link lists.

  • Product search
  • Product variants
  • Product display

Act and resolve

Get things done instead of redirecting.

  • Run process
  • Set variables
  • Send email
  • Rich messages

Steer the conversation

The right next step, on the phone too.

  • Return to the process
  • Transfer call
  • Hang up

Read complex documents in full

For long, nested documents such as insurance terms and conditions or technical manuals, classic knowledge search (RAG) only returns fragments and therefore no reliable answers. With the file tools, the agent finds the right document, reads it in full and answers from the complete context.

External tools via MCP

Your agent uses the tools of your own systems and partners via the Model Context Protocol: CRM, ERP, ticketing, appointment booking or customer account.

  • ✓You decide which tools an agent may use.
  • ✓Every call is traceable in the conversation.
  • ✓In addition: REST APIs via process modules.

More about MCP →

Build

Your business team builds the agent

No IT project needed: model processes visually in the LoyJoy Manager or describe the agent in Claude or ChatGPT. LoyJoy Headless creates the draft, you approve the release.

●
Visual.
No-code editor based on BPMN 2.0, with versioning and test mode.
●
Headless.
Create and change agents by prompt from Claude or ChatGPT. Included in all plans.

Discover LoyJoy Headless →

Resolve

Resolve the case instead of forwarding it

The agent does not just answer, it gets things done: meter readings, address changes, appointment booking, product advice or card blocking run all the way to completion.

●
100+ process modules.
Forms, authentication, appointment booking, product search, giveaways and more.
●
Integrations.
CRM, ERP and line-of-business systems via API or Model Context Protocol (MCP).

See process modules →Integrations →Model Context Protocol →

Deploy

One agent for chat and phone

Build the agent once and deploy it on website, app, messengers and phone. Knowledge, processes and analytics stay in one place.

●
Phone.
Natural real-time conversation instead of an IVR menu.
●
Chat.
Website, app, WhatsApp and other messengers.

See the phone agent →WhatsApp →

Control

You stay in control

Guardrails for production use in regulated industries.

Approved knowledge only

The agent answers from your sources only. Every answer can be traced back to its source.

Free choice of model

Leading models hosted in the EU, or an open-source LLM on LoyJoy-owned hardware in a data center in Germany. No training on your data.

Learn more

People take over

For edge cases, the agent hands over to your team with full context, via live chat or callback.

Learn more

Which requests will your agents resolve next month?

In 30 minutes, we show you which customer requests your agents can resolve starting next month, and how you measure the result.