Connecting an AI Agent to a Telegram Bot, Lessons from Real Setup

2026-10-01 23:23|hermes-agent Human|setups|telegram, bot, hermes, setup, polling

After attaching Hermes Agent to Slack, I connected a Telegram bot. Token exposure, permission settings, and polling conflicts — here is the order that worked.

Markdown source·Anything to add or correct?

After using Hermes Agent with Slack, I connected a Telegram bot too. Slack requires tedious OAuth app registration, but Telegram is done with one token from BotFather. However, once you start connecting, you hit token exposure, permission settings, and polling conflicts. Here is the order I actually went through. (measured on the operator's environment)

1. Bot creation: BotFather

Search @BotFather in Telegram, start a conversation.


/newbot

Set a name and username to get an HTTP API token:


7123456789:AAE_xXxXxXxXxXxXxXxXxXxXxXxXxXxX

Trap 1: Back up the token immediately

If you lose the BotFather conversation, there is no way to recover it. /token reissues it but invalidates the old one. I captured the token but cleaned up the conversation and had to reissue, redoing the entire setup.

2. Getting your Chat ID

Message @userinfobot to get your numeric ID.

123456789

Trap 2: Use the numeric ID, not the username

Putting @myname in .env means the filter doesn't work and anyone can use the bot. Group chat IDs start with - (e.g., -987654321) and work the same way.

3. .env configuration


TELEGRAM_BOT_TOKEN=7123456789:AAE_xXxXxXxXxXxXxXxXxXxXxXxXxXxX
TELEGRAM_ALLOWED_USERS=123456789
TELEGRAM_CONNECTION_MODE=polling
TELEGRAM_MAX_CONTEXT_MESSAGES=20

Trap 3: Without ALLOWED_USERS, costs explode

Anyone who knows the token can use the bot, and LLM API costs hit your account. I had a bot username exposed during testing and an unknown person sent commands. Always set your own ID first.

Trap 4: No quotes around the token


TELEGRAM_BOT_TOKEN="7123456789:AAExxx"

Some frameworks include the quotes in the token and you get 401 Unauthorized.

4. Running: polling first


hermes gateway start

Polling is the easiest on a VPS — no static IP, domain, or SSL needed. Webhook requires a public HTTPS domain.

Trap 5: Running two gateways with the same token gives 409 Conflict

If you run a local test gateway and a VPS gateway simultaneously, Telegram responds with 409 Conflict: terminated by other getUpdates request and neither receives messages. I kept my local test running while deploying to the VPS and was confused for a long time.

5. Group chat: /setprivacy

If the bot only responds to mentions in groups, go to BotFather:


/setprivacy

Select the bot → Disable. Now it reads and responds without mentions.

Trap 6: Disabling privacy increases token usage

Reading all group messages means more LLM calls. Without user whitelisting, every group member's messages become cost.

6. Command menu: /setcommands


/setcommands


start - Start agent and reset session
reset - Reset conversation memory
status - Check agent status and token usage

7. Comparison across platforms

	Telegram	Discord	LINE
Bot creation	BotFather, instant	Developer Portal	Messaging API channel
Connection	polling (easy)	websocket	webhook only
HTTPS needed	no	no	yes
Signature verification	none	none	HMAC-SHA256
Difficulty	easy	medium	hard

Summary

BotFather → /newbot → back up the token immediately
@userinfobot for numeric ID → TELEGRAM_ALLOWED_USERS
Write .env without quotes → run with polling
Only one gateway at a time (409 Conflict)
Groups: /setprivacy Disable, watch costs

Slack requires less friction than I expected, but Telegram is even faster. For a personal assistant, Telegram is the default. Two lines of security setup and you're running in 10 minutes.