Hermes Agent Bot Mode Explained: What It Is, How To Set It Up, Every Feature And Which Model To Pin To Each Bot

Hermes Agent Bot Mode is a free feature of the Hermes desktop app that turns each Hermes profile into a named bot with its own chat, model, memory, skills, avatar and schedule.

Nous Research made it, it is MIT licensed, and the official Bot Mode docs say it now ships built into the desktop app and is on by default.

Below you'll find what it is, who made it, what it costs, the setup, how to create a bot, every feature, the limits, my Grok Bot comparison and a late section on which model suits each bot.

I installed Bot Mode on launch day in August 2026 and documented that install in my Hermes Bot Mode guide on agentos.guide.

Anything newer than that build comes from the official docs, which I re-read on 12 October 2026, and I'll say which is which.

What Hermes Agent Bot Mode is in plain English

Hermes Agent is the free, open-source AI agent from Nous Research.

A Hermes profile is one isolated agent setup with its own config, memory, skills, credentials and chat history in its own folder.

Bot Mode is an interface on top of profiles that shows each one as a named bot in a roster.

You click a bot and you land in its chat, the same way you'd message someone on your team.

The docs say a bot is a profile, so there is no new primitive to learn.

The docs also say Bot Mode adds no core patches, no background daemons and no extra storage.

Who made Bot Mode and what it costs

Nous Research made Bot Mode, and Teknium from the Nous team shipped the first version as a public beta plugin.

The plugin lived in the NousResearch/Hermes-Bot-Mode repository under an MIT licence.

GitHub shows that repository was created on 13 August 2026.

Somebody told Teknium it looked a lot like Grok Bot, and he replied, "That's the point."

The repository is now archived, and its README says Bot Mode became a bundled, default-on desktop plugin through pull request #87886.

GitHub shows that pull request was merged on 17 August 2026.

The software costs nothing.

Your only cost is the model usage behind each bot, and a bot on a local model has no model bill.

The problem Bot Mode solves

Before Bot Mode, Hermes worked in sessions.

My inbox summary lived in one session, my content research in another and my competitor tracking in a third.

They looked identical from the outside, which was a long list of chats with no names and no jobs.

I was the routing layer between them.

That table is my summary of the comparison in the official docs.

How to set up Hermes Agent Bot Mode

The setup is short now, because the feature is built in.

1. Install Hermes Agent with curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash if you don't have it.
2. Run hermes setup and connect a model provider.
3. Run hermes update so you're on a current build.
4. Launch the desktop app, which the docs say you can do with hermes desktop.
5. Click the Bots tab next to Sessions in the left sidebar.
6. Check that a Routines tile is docked beside the conversation.

The docs say the desktop app runs on macOS with Apple Silicon, Windows and Linux.

Your existing profiles appear as bots straight away, so you rebuild nothing.

Older tutorials tell you to clone the plugin into ~/.hermes/desktop-plugins/hermes-bots first.

The archived README says no manual install is needed on current desktop builds.

How to create a bot

You press New Agent and fill in three fields, which are Name, Title and Description.

In my guide I typed "scout", then "Competitor scout", then one line about its job, and it took about twenty seconds.

A real profile appeared on disk with its own memory, skills and inbox.

The docs say a new bot introduces itself as the first message of its own chat.

An Advanced section opens the full set of options.

The docs flag one catch, which is that single-use OAuth logins for Anthropic, OpenAI Codex and xAI are not copied to a new bot.

You sign that bot in with hermes -p <name> auth add <provider>.

Every Hermes Agent Bot Mode feature

Routines

A routine is a recurring task that lives next to the bot that does it.

In my guide I gave my inbox bot an "Evening wrap-up" routine at 6pm, and Hermes cron returned the next run time.

Routines are cron jobs named [bot:<name>] <routine>, so they also show in hermes cron list.

The result lands in that bot's chat, so the summary is waiting when you click the bot.

The Agent Inbox, @mentions and bot-to-bot messages

The Agent Inbox was the launch-day name for a bot's persistent conversation where messages from other agents land.

The current docs call each bot's permanent conversation Bot Chat, and agent messages are delivered into it.

You type @researcher have a look at this in any chat.

The active bot hands the work off, and the reply comes back in the same conversation with the sender named.

The docs say the bot composes its own message with a tool called message_agent, so your words are not forwarded verbatim.

Every message carries attribution, which the docs show as "Message from 🤖 researcher (@researcher):".

An early tester named Devin Davis said he'd spent a week trying to coordinate two profiles with a kanban board, and he had this running in one morning.

Memory, SOUL.md and cloning

Each bot keeps its own memory, and Teknium confirmed that at launch.

Every bot has a SOUL.md file that defines how it talks, what it cares about and what it never does.

Duplicate copies everything, including the memory, so a trained bot can seed the next project.

Group chats

Group chats did not exist on launch day, and the answer then was "not yet".

The docs now describe rooms where each bot replies briefly or passes.

A bot can escalate a judgement call to you with @user, and the room shows a "needs you" badge.

Hard caps of ten messages per send and three rounds keep a room from spinning.

I haven't documented group chats in my own guide yet, so this part is from the docs.

How I wired Bot Mode into my Agent OS

I run an Agent Operating System, which is one dashboard for every agent I use.

I rebuilt the Bot Mode roster inside it as a Bots tab in the Hermes section.

That tab reads the same Hermes profiles the desktop app reads, so there is one team and one source of truth.

My guide shows 46 bots in that roster, and each one is a real profile on my machine.

Oracle watches my competitors, Astros pulls keyword ideas out of search data and Apollo handles voice.

Those were separate profiles I hopped between, and now one can message another without me copying between windows.

Limits and gotchas

• Bot Mode lives in the desktop app, so a headless server on its own won't show a Bots tab.
• The docs say the message_agent tool doesn't appear on a headless install until you add two markers by hand.
• Bot-to-bot delivery is per-invocation, so the receiving bot reads a message when it next runs.
• Live interrupt of a bot in the middle of a conversation is listed as future work.
• The desktop keeps three local bot backends warm by default, and the docs put each at about 60 MB.
• A bot opened while every slot is busy waits up to 30 seconds, so a large team needs Warm Bot Backends raised.
• A bot with no routine sits idle until somebody messages it.
• You can switch Bot Mode off in the desktop app's plugin settings, and the docs say your profiles, sessions and cron jobs stay untouched.

A viewer told me the interface looked a bit ugly next to Grok Bot at launch, and that was fair for a one-day beta.

Hermes Agent Bot Mode vs Grok Bot

This is the comparison from my video and my Hermes Bot vs Grok Bot post, which I made in August 2026.

Grok Bot isn't bad, and it showed the world what a team of named agents looks like.

One tester in the community had Hermes rebuild their whole Grok Bot team inside Bot Mode.

Grok Bot may have changed since August, so check its current page before you choose.

Which model should you pin to each bot?

Bot Mode doesn't ship a model of its own.

Every bot runs on whatever model you pin to it, or on the model of the launch profile when you leave the pin empty.

That makes model choice the one decision in Bot Mode that changes both quality and cost.

GoldieBench scores models on one-shot builds, which is not the same job as running a bot all day.

I'd still use the board as a first filter, because a model that can't finish a build won't hold up in a long bot conversation either.

Those picks are my suggestions from the board, and they are not a list of the models behind my own bots.

In my guide I describe the pattern I use, which is one model for research, another for writing and a small local model for the inbox bot.

Grok 4.7 is worth a note, because it has only 20 scored tasks, so treat its average as a smaller sample.

You can pin an xAI model to a single Hermes bot, and my August comparison found Grok Bot ran on xAI models only.

Local models live on the local board, and my best local model for Hermes Agent post goes deeper.

My best Hermes Agent model ranking covers the full list, and my best LLMs for Hermes Agent post covers the wider field.

I haven't benchmarked a model inside Bot Mode itself, so every number above is the model's one-shot build average and nothing more.

Related posts on GoldieBench

My Hermes Agent bot post covers Hermes as a bot on Telegram, Discord and Slack.

My Hermes Agent background post covers cron, background tasks and the gateway service.

Also On Our Network

• Agent Operatorsthe full Bot Mode setup walkthrough for operators
• AI Income Deskhow a business would use a Hermes bot team
• AI Tool Verdictevery Bot Mode feature explained and scored
• agentos.guidethe Bot Mode quick-start and how it fits an Agent OS
• agentos.guideMy original launch-day Hermes Bot Mode guide on agentos.guide

Create one bot, pin one model to it and give it one routine, and you'll learn more in a day than any table can tell you about Hermes Agent Bot Mode.