How MuseRelay works — Channels, Chatbots and AI Agents

ES EN

This guide explains the mental model of MuseRelay: the 3 building blocks (Channels, Chatbots and AI Agents) and how they fit together so your business can handle conversations — with humans, with AI, or both at once.

In one sentence: a channel or a chatbot is the entry point for messages. Conversations is where you read and reply to them. If you want AI to answer automatically, you assign it an AI Agent (the "brain"). You can pause it anytime to step in yourself.

🧩 The 3 core concepts

1. Channels — WhatsApp, Instagram, Messenger

A channel is a connection to an external messaging network. You receive messages from there and reply through there.

Available types today:

WhatsApp Business (Meta Cloud API)
Instagram Direct Messages
Facebook Messenger
Email, SMS and Telegram — coming soon

Channels are created from Management → Channels in the top menu. You need credentials from your own Meta account (BYOA model) — the platform guides you step by step.

2. Chatbots — widget embedded on your website

A chatbot is the chat bubble you place on your website. Your visitors click it and chat with you (or with the AI) directly from the browser, without installing anything.

Created from Management → Chatbots. You get a <script> snippet to paste into your website's HTML. Configurable: color, position, initial greeting, logo, welcome message.

Key difference: a channel connects to an external network (Meta). A chatbot lives on your own website. Both end up in the same place: your conversations.

3. AI Agents — the "brain" that replies

An AI Agent is a configuration that tells the AI model how to behave: what personality to have, what it knows about your company, what it can and cannot do, what tools it can use (like booking appointments in Cal.com).

Created from Management → AI Agents. Each agent has:

A name and description.
A prompt that defines its behavior (you can use dynamic variables like the current date, customer name, etc.).
A configured AI model (the system picks it — you only care that it's "Data center: Europe").
A token limit per reply.
Optionally, tools like Cal.com so it can book real appointments.

🔗 How the 3 connect

This is the key idea:

Channel (WhatsApp/IG/Messenger)
    │
    │   or
    │
Chatbot (bubble on your website)
    │
    │
    ▼
┌────────────────────────┐
│   Conversations        │  ← everything lands here
└────────────────────────┘
    │
    │   (optional)
    ▼
Assigned AI Agent ──────► replies automatically
    │
    │   (you can pause it anytime)
    ▼
You (human) ──────────────► reply manually

In other words:

Channels and chatbots are message entry points.
Everything lands in Conversations.
If you have assigned an AI Agent to the channel/chatbot, that agent replies automatically.
If there's no agent assigned (or you've paused it), a human (you or your team) replies from the interface.

👥 Human, AI, or both?

MuseRelay supports all 3 modes. Pick based on your case:

🙋 Humans only

Don't assign any AI Agent. All conversations arrive in Conversations and your team replies manually. Useful for small businesses with low volume or sectors where AI isn't appropriate.

🤖 AI only

Assign an AI Agent to the channel/chatbot and leave it active 24/7. The agent answers everything in real time. Useful for FAQ, simple bookings, lead qualification.

🔁 Hybrid (most common)

AI answers by default, but any human agent can pause the AI in a specific conversation to step in. When finished, reactivate the AI. Ideal setup: AI handles 80% and you handle the 20% that needs a human touch.

📍 Where conversations are handled

Two interfaces to manage conversations depending on your case:

Interface	URL	When to use it
Conversations	`/conversations`	Human intervention in any channel or chatbot. See all messages and reply personally.
Agent Panel	`/agent/conversations`	Dedicated view for customer support agents. Includes real-time notifications, conversation assignment and team tools.

If you work solo, /conversations is enough. If you have a support team, the Agent Panel is more convenient.

🚀 Recommended steps to get started

Create an AI Agent first (the brain).
- Go to Management → AI Agents → Create.
- Give it a name, write the behavior prompt.
- Save it and test in the playground (right panel of the edit view).
Create a Chatbot or a Channel (the entry point).
- For your website: Management → Chatbots → Create and copy the <script>.
- For WhatsApp/Instagram/Messenger: Management → Channels → Create and follow the wizard.
Assign the AI Agent to the chatbot or channel.
- From the chatbot/channel edit view, pick the agent from the selector.
Test by sending a message from your website or WhatsApp.
If you need to step in manually, go to Conversations, pause the AI and reply yourself.

💡 Advanced concepts

One AI Agent can be assigned to several channels/chatbots. No need to create a different one for each — if your business speaks the same way on WhatsApp, Instagram and web, a single agent works for all three.
Conversations are preserved even if you pause the AI. Full history stays in /conversations for future reference.
You can have channels without AI. If you prefer everything to go through a human first, simply don't assign an agent to the channel.
Variables in the agent prompt (date, customer name, etc.) are substituted automatically in real time. See Prompt variables.

❓ FAQ

Can I have a chatbot on my website without AI?

Yes. Create the chatbot, don't assign an AI Agent. Messages arrive at /conversations and you reply manually.

Does AI keep replying if I pause a conversation?

No. When you pause the AI in a specific conversation, all following messages wait for a human response. When reactivated, it replies again.

Can I use different AI agents for different channels?

Yes. For example, a "customer support" agent on WhatsApp and a "sales" agent on the landing chatbot.

Where do I configure channel-specific details?

Each channel type has its own documentation with detailed steps (Meta credentials, webhooks, etc.). Check the Channels section in the left sidebar of the documentation.