Frequently asked questions

This page answers the questions that come up most often in conversations about msg.Hello. Each answer stands on its own; the links lead to the detailed chapter.

​

Basics

​

What does msg.Hello do?

msg.Hello answers calls for you. An agent greets the caller, understands their request, completes the task through your connected systems, and closes the conversation – by hanging up or forwarding to a human.

​

Is this just a phone menu with a better voice?

No. A phone menu asks for digits; an agent holds a conversation and completes real tasks along the way. It understands free phrasing, asks when something is missing, and calls tools in your business system.

​

Which requests is an agent suited for?

Best for recurring calls with a clear course: appointments, information, identification, status requests, simple agreements. The clearer the course, the more reliable the agent.

​

Is msg.Hello tied to one industry?

No. The conversation course is built per agent, and the business logic comes from your own system via connectors or an MCP server. The platform provides the structure, not the industry.

​

Setup

​

How long does it take to get the first agent?

A simple agent that recognises requests and forwards them takes just a few steps – see Getting started. The time goes not into creating it, but into thinking the conversation through and connecting the business systems.

​

Do I need an interface to my system?

Only if the agent is meant to do something there. Greeting, sorting, forwarding, and hanging up work without any connection at all. For business tasks you need a connector or an MCP server.

​

Who can configure an agent?

Anyone with the appropriate role in your tenant – these are business roles, not developer roles. You control who may do what under Users & team.

​

Do I need to be able to program?

No. Flow, instructions, and tools are configured in the portal. You need programming skills only on the other side of the interface, when your business system is to be connected.

​

Where does the phone number come from?

The phone number is set when the agent is created and must be registered with our provider. A new agent starts without a phone number and without a live version. Alternatively, your own phone system forwards calls to the agent via SIP, see Telephony connection.

​

Can an agent also place calls itself?

Yes, via the API. An agent with a live version and a phone number calls a number and conducts the conversation according to its flow. Details under Outbound calls.

​

Do I have to build every agent from scratch?

No. When creating an agent, you can choose a template instead of an empty agent. It already includes instructions, connectors and flow. You only enter the credentials yourself.

​

Does the agent know the caller’s number?

Yes, if you allow it. With the built-in tool get_caller_phone_number, the agent reads the transmitted phone number. If the number is withheld, it receives the code ANONYMOUS and can ask for it.

​

In operation

​

Can I change something while the agent is live?

Yes. You always work on the draft; this has no effect on ongoing calls. Only when you publish and activate a version do new calls use the new state – ongoing calls stay on their version.

​

How do I check a change without calling?

With the test call in the browser. It runs against your current draft and shows the state, the conversation as text, and every tool call live – and it creates no entry in the call analysis.

​

What happens if the agent gets stuck?

It stays in the current state and can ask again. For the case where things do not move forward, you build an exit: forward to a human, or end the conversation in an orderly way.

​

What happens if my business system fails?

The tool call reports an error result – for MCP tools, the response code ERROR. You decide in the flow what happens then; forwarding to a human is the usual choice. A failure does not lead to the agent making things up.

​

Can the caller interrupt the agent?

Yes, if you enable barge-in for the agent. Together with noise suppression, voice, and transcription language, this is part of the model parameters under Manage agents.

​

How long may a conversation last?

You define this per agent: maximum waiting time for an input and maximum call duration. Both are unlimited by default and are enabled with a switch.

​

Control & security

​

Can an agent promise something it is not allowed to?

No – provided you assign the tools cleanly per state. An agent can only do what a tool has been permitted for in the current state; everything else does not exist for it at that point. The reasoning is explained under Concepts & Control.

​

Can a caller persuade the agent to skip a step?

No. A state switches only when a tool that was actually executed reports a result for which you have defined a transition. Without that result, the agent stays where it is – along with the tools available there.

​

Are conversations recorded?

Calls and transcripts are available after the conversation in the call analysis. How long they are retained is set in the settings. Test calls are not logged.

​

Is old data deleted automatically?

The retention periods for calls and dialogues are already configurable. The automatic, time-controlled deletion after a period expires is currently being finalised.

​

How do users sign in?

Through your organisation’s single sign-on. Passwords and sign-in procedures are managed by your identity provider, not by msg.Hello.

​

How safe is connecting a third-party MCP server?

Treat it as an external trust boundary: for each agent, permit only the tools it genuinely needs, and check the behaviour with a test call beforehand. Credentials are stored encrypted. Instead of a permanent token, you can also store OAuth client credentials. The agent then works only with short-lived access tokens.