Hermes Agent Bot Explained: What It Is, How To Set One Up On Telegram And Which Models To Run Behind It

A Hermes Agent bot is your Hermes Agent running inside a chat app such as Telegram, Discord, Slack or WhatsApp, so you can message it from any device.

It works through the messaging gateway that is built into Hermes Agent, the free, open-source agent from Nous Research.

Below you'll find what it is, every platform, the Telegram setup step by step, group behaviour, approvals, costs and a late section on which models to run behind it.

I checked every command against the official Hermes messaging docs and against the help text on my own install, which is version 0.21.4.

I drive most of my Hermes work from the Bots tab in my Agent OS, so the Telegram steps come from the docs and the CLI help and not from a recorded test of mine.

What a Hermes Agent bot is in plain English

Hermes Agent is an AI agent that can use tools, run commands, remember things and work on a schedule.

Most people use it in a terminal or in the desktop app.

A Hermes Agent bot is that same agent with a chat app as its front door.

The docs describe the gateway as a single background process that connects to all your configured platforms, handles sessions, runs cron jobs and delivers voice messages.

Each message goes through the full agent, with its tools, memory and skills.

That is the difference between this and a simple reply bot.

Who makes it and what it costs

Nous Research makes Hermes Agent, and the code is open source under the MIT licence.

The gateway is part of that free install.

Creating a Telegram bot through BotFather is also free.

The model you connect is the part that can cost money, and I cover that in the cost section and the model section below.

Which platforms a Hermes Agent bot runs on

The docs list Telegram, Discord, Slack, WhatsApp, Signal, SMS, Email, Home Assistant, Mattermost, Matrix, DingTalk, Feishu/Lark, WeCom, Weixin, BlueBubbles for iMessage, QQ, Yuanbao, Microsoft Teams, LINE, ntfy and your browser.

The docs also publish a feature table, and these are the rows most people need.

I'd start with Telegram, because it needs one free token and no public server.

The built-in WhatsApp bridge is not the official Business API, so the docs recommend a dedicated phone number.

The docs describe a second WhatsApp route through the Business Cloud API, which needs a Meta Business account and a public webhook.

How to create a Hermes Agent bot on Telegram

You need Hermes installed and a model connected before you start.

Step 1: Create the bot in BotFather

Open Telegram and message @BotFather, which is Telegram's official bot management tool.

Send /newbot, choose a display name and choose a username that ends in "bot".

BotFather replies with an API token.

Step 2: Protect the token

Anyone who has the token can control your bot.

If it leaks, you revoke it with /revoke in BotFather.

Step 3: Get your user ID

Hermes controls access with numeric Telegram user IDs.

Message @userinfobot and it replies with your number.

Step 4: Run the wizard

Run hermes gateway setup and choose Telegram.

The wizard asks for the bot token and your allowed user IDs, then writes the configuration.

The manual route is to add TELEGRAM_BOT_TOKEN and TELEGRAM_ALLOWED_USERS to the .env file in your Hermes home folder.

The docs also say the Messaging page in the dashboard and the desktop app has a "Create with QR" button for Telegram.

Step 5: Test it

Run hermes gateway to start the gateway in the foreground.

Send the bot a message, and the docs say it should come online within seconds.

Step 6: Run it as a service

Run hermes gateway install and then hermes gateway start.

That installs a systemd service on Linux or a launchd service on a Mac.

Run hermes gateway status to check it.

On a server that should survive a reboot, the docs give sudo hermes gateway install --system.

Step 7: Set a home channel

Type /sethome in the chat where scheduled results should arrive.

A job created with hermes cron create and the --deliver telegram flag then lands in that chat.

How the bot behaves in DMs and groups

In a direct message, the bot answers every message from an allowed user.

In a Telegram group, privacy mode decides what the bot can see.

Privacy mode is on for every new bot, so the bot only sees slash commands, replies to its own messages and service messages.

You turn it off by sending /mybots to BotFather and choosing Bot Settings, Group Privacy and Turn off.

You then remove the bot from the group and add it again, because Telegram caches the setting.

Promoting the bot to group admin has the same effect.

Your allowlist still applies inside a group.

Setting telegram.require_mention: true keeps the bot quiet until someone tags it or replies to it.

On Discord, Hermes answers every DM and only answers in server channels when it is mentioned.

If several Hermes bots share a group, the docs say each profile needs its own bot token.

Approvals and safety on a Hermes Agent bot

The docs say the gateway denies all users who aren't in an allowlist or paired through a direct message.

• You can add numeric user IDs to the allowlist for each platform.
• You can let an unknown user receive a one-time pairing code, and approve it with hermes pairing approve telegram followed by the code.
• Pairing codes expire after one hour, and hermes pairing revoke removes access.
• You can make yourself an admin and limit other users to the slash commands you choose.

Hermes also checks commands against a list of dangerous patterns before it runs them.

The default approval mode is smart, which approves low-risk commands, denies dangerous ones and asks you about the uncertain ones.

You answer in the chat with /approve or /deny.

Scheduled jobs deny dangerous commands by default, because nobody is there to answer.

The command hermes pause is a global emergency stop for new work until you run hermes resume.

The chat commands worth knowing

Sessions don't reset after a quiet period, so the bot keeps its context until you type /new.

Voice notes on Telegram are transcribed automatically, and the local option uses faster-whisper with no API key.

What a Hermes Agent bot costs

The docs note that bots need a model provider and tool providers for speech and web search, and that a Nous Portal subscription bundles them.

Free model lists rotate, so I don't build an important process on one free model.

My sister post on the best Hermes Agent VPS covers the always-on machine.

Limits and gotchas

• A cold boot drops Telegram messages that arrived while the gateway was offline, unless you set drop_pending_on_cold_boot to false.
• Two running gateways can't poll the same Telegram bot token.
• Hermes uses long polling by default, and the docs describe a webhook mode for cloud machines that sleep when idle.
• The WhatsApp bridge can stop working after a WhatsApp update, and the docs say to update Hermes and re-pair.
• Files are sent by the gateway process, so a file that only exists inside a Docker container can't be delivered until you mount a shared folder.

Bot Mode and Hermes Bots are a separate feature

The word "bot" also describes a different Hermes feature.

Hermes Bot Mode is a desktop plugin from Nous Research that shows your Hermes profiles as a roster of named bots.

Each bot has its own chat, memory and routines, and you can pin a different model to each one.

Bots message each other through the Agent Inbox.

I rebuilt that roster as a Bots tab in my own Agent OS, where creating a bot takes a name, a title and a description.

In my latest Hermes video I also mention creating a Hermes bot from an active profile, so a profile I use for one model becomes a named bot.

My post on Hermes Agent Bot Mode covers that feature on its own, and my Hermes Bot Mode guide on agentos.guide shows it running.

The gateway service itself is covered in my post on Hermes Agent background work.

Which model should sit behind a Hermes Agent bot?

This is the part of the decision that GoldieBench can help with.

Hermes Agent doesn't ship its own model, so a bot is only as good as the brain you connect.

GoldieBench gives every model the same one-shot prompt to build a game, a page, a simulation or a visual, then scores the result from 0 to 10.

It doesn't measure chat replies, tool calling over Telegram or uptime.

That means these numbers are a signal about how capable the brain is, and they aren't a score for the bot.

The gap between the top and the bottom of that table is the lesson.

A bot on Claude Opus 5 and a bot on a small free model are the same software with very different results on hard work.

For a morning brief or a quick lookup, a cheap or free model is often enough.

For a job where the bot builds something, I'd put a model from the top of the table behind it.

Free models and what the board can't tell you

The free list on Nous Portal changes often.

When I recorded my MissingNo video, it showed about eight free models, including Step 3.7 Flash, LongCat, Laguna XS 2.1, Solar Mini and MissingNo.

MissingNo, Step 3.7 Flash and Solar Mini don't have pages on the board, so I can't give them a score.

LongCat-2.0 has a page with a provisional average of 8.12, and it is provisional because only four tasks are scored.

I wouldn't rank a model on four tasks, so treat that number as early.

One gateway can run different brains

The docs describe per-channel overrides, where a single gateway runs a different model in each channel.

Their example is a cheap, fast model in a daily channel and a stronger model in a development channel.

You can also change the model inside a chat with the /model command.

That lets you match the score to the job rather than paying frontier prices for every message.

How Hermes does with strong models

One entry on the board shows what the Hermes harness can do with strong brains behind it.

That run was done on my desktop Hermes install and not through a Telegram bot.

My best Hermes Agent model ranking and my best LLMs for Hermes Agent post go deeper on the brains.

If you'd rather run a free brain on your own machine, the local models board shows what runs offline.

Also On Our Network

• Agent Operatorsthe full Hermes Telegram bot setup for operators
• AI Income Deskeight ways a business uses a Hermes bot
• AI Tool Verdictevery Hermes bot platform and feature, scored
• agentos.guidethe Hermes bot quick-start and how it fits an Agent OS
• aiprofitboardroom.comWhy I moved my agent team onto Hermes bots

Create the token, lock the allowlist, pick a brain that matches the job, and you'll have a useful Hermes Agent bot.