You can write to an agent from Telegram, through your own bot: it answers you there, with the same tools and the same limits as in a conversation. With triggers, an agent can also start on its own when an event happens in one of your services.
TelegramLink to the Telegram section
What to knowLink to the What to know section
- You create your own Telegram bot. The Telegram account that sends the link code becomes the only one that can talk to the agent: anyone else is ignored.
- The conversations also appear in Moon AI.
- Messages pass in plain text through Telegram and through our connection service, in the United States, which keeps them for up to a year. Moon Blur only protects them from the AI providers.
1. Create your botLink to the 1. Create your bot section
- In Telegram, open @BotFather and send
/newbot. - Choose a name, then a username that ends in
bot, for exampleemma_moon_bot. - Copy the token BotFather gives you.
- Recommended: send
/to BotFather, then “Disable”, so that nobody can add your bot to a group.setjoingroups
Use a bot created for Moon AI only: a bot already used by another service stops receiving its messages there.
2. Connect TelegramLink to the 2. Connect Telegram section
- “Connectors” › “Catalog” › Telegram › “Connect”.
- Paste the bot token on the connection page, then come back to Moon AI.
3. Link the bot to an agentLink to the 3. Link the bot to an agent section
- “Connectors” › “Messaging apps” › “Link a messaging app”, or, on an agent’s page, “Talk to it from a messaging app”.
- Choose the “Agent that answers”.
- Set the options (see below).
- Paste the bot token again when Moon AI asks for it. Moon AI keeps it encrypted and only uses it to set up your bot, show that it is typing and stop it.
- Check the consent box, then click “Link app”.
- Open the link on the phone where Telegram is installed, or scan the QR code: Telegram opens your bot. Tap “Start”.
The link code is valid for 10 minutes and allows 5 attempts. If it has expired, open the link’s details and click “Get a code”, then send it right away. The bot replies “Linked to Moon AI”, with the name of the default agent.
Link optionsLink to the Link options section
| Option | Effect |
|---|---|
| “Let the agent prepare actions” | The agent can prepare messages, posts, events and more. Each action waits for your approval, with a button in Telegram or in Moon AI. |
| “Follow my "Act without asking" settings” | Not recommended. Actions of connectors set to act without asking then run without approval, even from Telegram. |
| “Answer strangers” | Anyone else who writes to the bot gets “This bot is private.”, at most once a day. Otherwise, nobody else gets a reply. |
Talk to the agentLink to the Talk to the agent section
Write to the bot as you would to anyone. The bot shows that it is typing and replies “Working on it…” while the agent works.
| Command | Effect |
|---|---|
/new | Starts a new conversation. |
/agent | Lists the available agents. Reply /agent followed by the number to pick one for this conversation. |
/stop | Stops the current answer. |
/unlink | Unlinks the messaging app. Moon AI asks you to confirm with /. |
/help | Shows the help. |
The French commands (/nouveau, /aide, /delier) work too.
- Only text messages are read, in the private chat with the bot, 8,000 characters at most. A file sent without text is not supported.
- A forwarded message is treated as content written by someone else: the agent does not follow its instructions.
- A long answer ends with “Read the rest in Moon AI” and a link. Created files open in Moon AI.
- The “Tools used” line tells you which tools the agent called.
Approve an action from TelegramLink to the Approve an action from Telegram section
When the agent wants to act, Telegram shows the action with the “Approve” and “Decline” buttons, and a “See and approve in Moon AI” link. The same request also waits in “Pending approvals”. The first decision counts, wherever it is made. See Approve actions.
LimitsLink to the Limits section
| Limit | Value |
|---|---|
| Messages | 20 per 10 minutes and 300 per day, per link |
| Answers at the same time | 3 |
| Message length | 8,000 characters |
Each answer counts toward your usage limit, like a message in Moon AI. When the limit is reached, the bot says so, with the time it becomes available again.
Manage and unlinkLink to the Manage and unlink section
“Connectors” › “Messaging apps” lists your links, with the date of the last message. A link’s details show who can talk to the agent. From there you can change the “Default agent” (for new conversations; /agent picks another one within a conversation), “Pause” the link and “Resume” it, and “Unlink app”. If Moon AI needs the bot token again, the details ask for it: paste it, then “Save token”.
Once the messaging app is unlinked, the bot stops answering; your conversations stay in Moon AI. To use it again, link it with a new code. Deleting the agent that answers pauses the link until you choose another agent.
TriggersLink to the Triggers section
A trigger starts an agent on its own when something happens in one of your services: a new email, a comment, an upcoming meeting. The agent works in its own conversation and asks for your approval before acting.
Available eventsLink to the Available events section
| Service | Events |
|---|---|
| Gmail, Outlook | “New e-mail received” |
| Google Calendar | “Meeting starting soon” |
| Slack | “Message in a channel” |
| GitHub | “New comment”, “New issue”, “New pull request” |
| Notion | “New comment”, “New page” |
| Linear, Jira, Trello, Asana | “New comment” |
| YouTube | “New activity” |
The service must be connected in “Connectors” and not paused.
Create a triggerLink to the Create a trigger section
- Open the agent’s page, “Tasks” tab, “Triggers” section.
- Click “Add a trigger”.
- Choose the event under “When”.
- Fill in “Only these senders”: one address, domain or username per line. This is recommended, because anyone can send you an email or a comment.
- Write the “Instruction” under “What the agent does”, for example: “Summarize the email and draft a reply.”
- Choose “Where the result goes”: “In its conversation”, “In its conversation, with a notification”, or in its conversation and on a linked messaging app.
- Set “Maximum runs per day” (from 1 to 100).
- If needed, check “Let the agent prepare actions”: each action waits for your approval.
- Click “Create trigger”.
“Filters (optional)” adds “Ignore these senders” and “Only when it contains” (one word per line). An ignored event starts nothing and uses none of your limit. If the chosen service is paused, resume it first: it comes back everywhere, in conversations, agents and routines.
Follow and manageLink to the Follow and manage section
For each trigger, the “Triggers” section shows the number of runs today and the time of the last event. “Results” opens its conversation. “Edit”, “Pause”, “Resume” and “Delete trigger” manage it; deleting a trigger keeps its conversation in your history.
Each run uses your usage limit, like a message. You can have up to 20 triggers. If the page says that triggers are suspended for now, new events do not start the agent until they are back; your triggers keep their settings.