You turned on voice replies for one of your groups, waited, and... got another paragraph of text. Nothing is broken. Voice is a deliberately occasional extra — it is not the assistant's normal way of answering, and in two of the four modes most answers are still text.
This guide explains exactly what the setting does, what each mode means, when a voice note can appear, when the assistant always stays text, and how to switch it on.
It Replaces the Assistant's Own Reply
When voice replies are on, your assistant can answer with a real Telegram voice note instead of its own text reply. That is the whole trick: the answer it was already going to give is spoken instead of typed.
Voice replies never change whether your assistant answers. @mention rules, your group participation mode and occasional participation all work exactly as they did before — voice only changes the format of an answer the assistant had already decided to give.
The Four Modes
Voice replies are a single per-group choice with four modes:
- Off — your assistant always answers with text in this group. This is the default.
- Voice answers voice — a voice message gets a voice note back. Typed messages still get text.
- Occasional voice — mostly text. Now and then a conversational answer arrives as a voice note.
- Always voice — answers with a voice note whenever it fits. Long or technical answers stay text.
You can change the mode per group at any time, and each group keeps its own choice.
When It Always Stays Text
Some answers are never spoken. They always arrive as text:
- long answers;
- technical or structured answers — code, headings, lists, tables or JSON;
- commands;
- error or fallback messages;
- replies that carry a file.
This is on purpose. Those answers are meant to be read and copied, and a spoken list or a code block is worse than the text. Direct messages are always text too — voice replies are a Telegram group feature.
Why It Feels Rare — and That's on Purpose
In Occasional voice, the assistant is deliberately restrained. Roughly one eligible answer in three passes, there are at least five text answers between voice notes, and there is a ten-minute pause after each voice note.
Those limits are there so a group never turns into a stream of audio. The practical result is that most messages still get a normal text answer. Getting text instead of a voice note is expected — it does not mean the feature failed, and nothing was lost.
The Voice You Choose Sets the Language
Below the mode you also pick your assistant's voice. That choice is not just about how it sounds: your assistant answers in that voice's language. Pick a Spanish (Spain) voice and it answers in Spanish even when you write in English.
The Mini App shows this under the voice picker, so the promise you see is the language you get. If you want your group answered in a particular language, choose a voice for that language.
Paid Plans Only, and Off by Default
Voice replies are a paid-plan feature, and they are off by default. On the free plan, voice messages you send still get a text answer; turning voice replies on requires a paid plan. Each group is set separately, so you can have voice in one group and text in another.
How to Turn It On
The control lives with the other per-group settings. It is off by default, and it is set separately for each group.
Open the mini app. In the group, type /settings and open the Agent One mini app.
Pick the group. Choose the group you want to change from the Groups list.
Choose a mode under "Voice replies". Pick Off, Voice answers voice, Occasional voice or Always voice.
Choose the assistant voice. This is the voice your assistant speaks with, and its language is the language your assistant answers in for that group.
If the option is locked, it means your account is not on a paid plan. The setting shows a short "available on paid plans" note instead of quietly doing nothing.
The Short Version
Voice replies are optional, paid, per group, and off by default. You pick one of four modes and an assistant voice; the voice also sets the language your assistant answers in. Long, technical, command, error and file replies always stay text, direct messages are always text, and in Occasional voice most answers are still text — getting text instead of a voice note is normal, not a failure.