Hermes Bot Mode just launched, and it is free.
It turns your AI into a roster of named bots you can actually see.
Each one gets a face, a job, its own memory, and its own schedule.
They message each other while you watch.
And there is one setting buried in the advanced menu that changes everything — I'll show you that in a minute.
You install it as a plugin, the community sends feedback, then it ships inside the main Hermes desktop app for everyone.
When someone said this looks a lot like Grok Bot, Teknium answered with three words. "That's the point."
What Bot Mode actually does, how it compares to Grok Bot, and the setting almost nobody will notice.
Up until today, Hermes worked in sessions.
You open a chat. You give it a task. It does the work.
Want a different job done? You open another session.
Your inbox summary lives in one. Your content research in another. Your competitor tracking in a third.
And they all look the same.
Just a long list of chats. No faces. No names. No jobs.
The Named Agent Engine breaks that pile apart for good.
Creating a bot is three text boxes. Name, title, what it does.
You'll watch me do it in about twenty seconds, further down this page.
Each bot has its own avatar, personality, chat, memory and schedule.
If you already have Hermes profiles, they show up as bots the moment you install it — you rebuild nothing.
What you're watching: my real roster, live in the Agent OS. Clicking down the list — researcher, writer, inboxbot — each one opens its own inbox, its own routines, its own personality file. 46 bots, all of them real Hermes profiles on this machine.
Name, title, description. That bot now exists.
What you're watching: me typing "scout", "Competitor scout", and one line about its job — then a real profile appears on disk with its own memory, skills and inbox, and slides straight into the roster.
The same three boxes in the official desktop plugin, from the Nous Research repo. Underneath the fold is an Advanced section — that's where the setting lives.
Open Advanced and you can pin a specific model to that one bot.
Research bot on one model. Writing bot on another. Inbox bot on a small, fast, free model that never leaves your machine.
The Advanced panel from the official plugin: clone from another bot, pin a provider and model, write a custom personality file, skip the bundled skills. This is the screen most people will scroll straight past.
Grok Bot runs on xAI's models, in xAI's cloud, under xAI's rules.
plugin.js, and you can read every lineNot here. Bot Mode is the interface — you still choose the brain behind each bot.
Point one at the best model money can buy and another at a free one. Same roster.
Your research bot remembers what it researched. Your content bot remembers your style. They never blur together.
Each one builds its own history over months, like a team member who's been in the role a while.
Bot Mode dropped this morning, and this became the roadmap: which bots to create first, what job to give each one, and which model to pin to each.
Every bot has a persistent inbox, and messages arrive with attribution — the receiving bot knows exactly which agent sent it.
Its personality file teaches it the protocol, including how to reply.
What you're watching: the real handoff underneath, replayed at reading pace. First message ever — the inbox thread doesn't exist yet, so it gets created and named. Every message after that resumes the same thread. This is the actual session from my machine.
The bot you're talking to hands the message off, waits for the answer, and reports back in the same conversation.
What you're watching, live and unedited: I'm in writer's chat. I type "@researcher one line: what is the first bot a business owner should create?" — writer hands it off, researcher wakes up and actually runs, and its answer lands back in this chat. One conversation. Two agents did the work.
You type "@researcher pull the latest facts on this topic" — researcher digs with its own memory and its own model, sends the answer back, and your main bot folds it into the draft.
Devin Davis said he'd spent a whole week trying to get two Hermes profiles to coordinate using a kanban board.
He set up bot-to-bot coordination and had it running the same morning.
The coordination used to be duct tape. Now it's built in.
They took their existing setup — the coordinator bot and every worker bot under it — and had Hermes rebuild the same structure inside Bot Mode.
Same team layout, on free open-source software they fully control.
Summarise my inbox at 7am. Check my niche at noon. Pull content ideas every Monday.
You attach the routine to the bot whose job it is, and the results land in that bot's own chat.
What you're watching: I give inboxbot a routine called "Evening wrap-up", set it to 6pm, and describe the job in one sentence. It's scheduled for real — the next run time comes straight back from Hermes cron.
The same Routines tile in the official desktop plugin. So when you wake up and click your inbox bot, the morning summary is already sitting there waiting for you.
Profiles already existed — each one an isolated setup with its own config, memory, skills, credentials and history, in its own folder on your machine.
Bot Mode is a new interface layered on top. No core patches. No background daemons. No extra storage.
Plugins that hack the core of an app break every time the app updates. This one can't break that way — it only uses the standard surface Hermes already exposes.
It's the difference between a house built on the foundation and a treehouse nailed to the side.
The personality file defines how the bot talks, what it cares about, and how it behaves.
Cloning copies everything: config, skills, personality file, and even the memory.
# Research bot Digs up facts, sources and competitor moves. Reports findings back with links, never guesses. ## Who you are You are 'researcher'. You are one agent on a roster of named agents. Every agent has one job. Yours is finding things out and coming back with something checkable. You are calm, literal and fast. You do not pad answers. When you do not know, you say so and name what you would need in order to know. ## How you answer 1. Lead with the finding, not the process. 2. Every claim carries a source you can click. 3. If two sources disagree, say so and show both. 4. Never invent a number. An estimate is labelled. 5. Keep it under 200 words unless asked for depth. ## Agent Inbox You have a persistent chat titled "Agent Inbox" where agent-to-agent messages land. A message that starts with [Message from agent 'x'] came from another agent, not from the user. To reply to that agent, run: hermes -p <agent-name> chat --in ~ -c "Agent Inbox" \ -q "[Message from agent 'researcher'] your message" --in ~ -c "Agent Inbox" appends to that ONE named conversation, so the thread builds up over time instead of starting fresh on every message. If the agent has never been messaged, that session does not exist yet. Start one, then rename it: hermes -p <agent-name> sessions rename <id> "Agent Inbox" ## What you never do - You never answer outside your job. Hand it off instead. - You never speak for another agent. - You never fabricate a source to fill a gap. ## Handing off If a request is really a writing job, mention the writer. If it is a scheduling job, mention the bot that owns it. One agent, one job — that is the whole point of a roster.
A real personality file at production depth — identity, answering rules, the inbox protocol, and the handoff etiquette. Spent weeks training up a research bot that knows your business? Duplicate it, point the copy at a different project, and the new one starts with everything the original learned.
Upload your own image, generate an AI portrait, or give a bot a pixel pet that bounces beside it while it's busy.
Silly? A little. But when you've got six bots and three are mid-task, you can see at a glance who's working. The eyes moving means work is happening. It's a status light with personality.
Oracle watches my competitors. Astros pulls keyword ideas out of search data. Apollo handles voice — and before today, those lived as separate profiles I'd hop between.
Now they've got faces, chats and schedules side by side in one panel, and one can message another without me copying and pasting between windows.
If you run the Agent OS, it's a new Bots tab in the Hermes section — the roster, the inboxes and the routines you've been watching all the way down this page.
The whole point of Bot Mode is removing the technical part — creating a bot is three text boxes, and the install is copying one folder.
Then reload plugins in the desktop app, and a Bots tab appears next to Sessions. After this beta you won't even do that — it ships built into the app.
The people struggling with Hermes a year ago were writing config files by hand.
The people using it today are clicking a button that says New Agent. That gap keeps shrinking, and it shrinks in your favour.
Four rough edges, straight from the repo — none of them deal breakers.
That pace is the actual story of Hermes as a project — features don't sit on roadmaps for quarters.
They ship, get feedback, and land in the main app.
Wrong: "Tools change every week — why learn this one, it'll be replaced in a month."
Right: Grok Bot proved people want a roster of named agents with jobs. Days later a free, open version exists. The tool matters less than the pattern — and the pattern is now everywhere.
Wrong: "Agent teams are a developer thing."
Right: "Summarise my inbox every morning" is not a developer task. Neither is "watch my market" or "draft my newsletter from these notes." Those are business-owner tasks, and they're the exact examples this was built around.
Wrong: "I'll wait until things settle down."
Right: It isn't settling down. Learn to run a roster once — what jobs to define, how to write a personality, how to split work — and that skill transfers to every version of this that ships from here.
Members post their wins every day — agency owners, ecom founders, course creators, solo operators across 38 countries. Real businesses, real numbers, in their own words.
Read the 158-page wins doc →Asked why open source can't match the closed labs' computer-use agents, Teknium said Nous is working closely with a company called cua to build exactly that, open source.
And the version you'll be running will be one you own, not one you rent.
After this beta window it ships to everyone by default — and the people who set up their roster now will have bots with months of memory by the time everyone else installs it.
A roster with names and faces replaces a list of identical chats.
Each bot pins its own model — cheap jobs stay cheap, hard jobs get the best brain.
Every bot keeps its own memory, so it knows its job cold.
Bots message each other directly, with attribution, in their own inbox.
Routines run overnight and leave the answer in the bot's chat.
MIT licence, your machine, your models, your folders.
Bot Mode gives you the roster. The Boardroom gives you the year of work around it — which bots to create first for a business like yours, how to write each one's job description and personality file so it actually behaves, which model to pin to each, and the morning routines that land summaries in your chat before you've had coffee.
You'll see it in the daily tutorials with step-by-step videos, and you can bring your own roster to the four weekly coaching calls and get it reviewed live. You also get the Agent Operating System — the zip file, the video walkthrough and the 30-day roadmap — where Hermes plugs in alongside Claude, OpenClaw and Free Claude Code, with daily updates as we improve it.
Readers bookmark this page. Operators install their first bot tonight and wake up to a summary that wrote itself.
Get the Agent OS → Inside the AI Profit Boardroom · skool.com/ai-profit-labNames and faces. We've had agents for a while — but there's something different about clicking on Oracle instead of Session 47.
You start treating the work like a team instead of a tool. And once your work runs like a team, you stop asking whether AI agents are ready, and start asking what job you're giving the next one.