Hermes profiles have been powerful for a while. They can isolate memory, skills, credentials, models, sessions, and personality into completely separate agents. The problem is that power has mostly felt like infrastructure.
Hermes Bot Mode changes the mental model.
Instead of thinking about profiles as folders and config files, Bot Mode turns them into a roster of named agents inside Hermes Desktop. Each bot gets its own chat, personality, avatar, routines, and persistent identity. More importantly, those bots can actually talk to each other.
That sounds cosmetic until you look at what Nous Research actually built underneath it. This isn’t a pretend multi-agent dashboard sitting on top of one shared assistant. A bot is a real Hermes profile, with the same isolation Hermes already uses under the hood.
Teknium dropped Bot Mode as a public beta, and after digging through the release, the official repo, and how the plugin is wired into Hermes, I think this is one of the most important Desktop ideas Nous has shipped yet.
What Hermes Bot Mode Actually Is
Hermes Bot Mode is a desktop plugin for Hermes Agent. Once installed, it adds a Bots pane to Hermes Desktop and turns every profile into something that feels much closer to a persistent member of an AI team.
The basic roster shows an avatar, latest-message preview, and timestamp for each bot. Click one and you land in that bot’s chat. That alone sounds simple, but it fixes a surprisingly important UX problem: profiles stop feeling like environments you switch into and start feeling like agents you go talk to.
Creating one is also much cleaner than building a profile by hand. The New Agent flow starts with a name, title, and description. Open the Advanced section and you can clone an existing profile, pin a provider or model, write a custom SOUL.md, and choose whether to include bundled skills.
Right-click an existing bot and you can edit its avatar, title, description, skills, toolsets, model, and SOUL.md. You can also duplicate a bot, which performs a full clone including config, skills, personality, memory, and appearance.
That’s the first place Bot Mode clicks for me. It isn’t inventing a second agent system. It’s making the profile system Hermes already has dramatically easier to understand and operate.
The Bots Finally Feel Like Individuals
Nous could have stopped at a profile list and called it a day. Instead, Bot Mode leans into identity.
You can use one of the built-in geometric faces, upload your own image, or generate a portrait when an image backend is configured. The geometric avatars have blinking eyes that scan while the bot is working. There’s even a pixel pet companion that bounces beside the avatar during activity.
Is any of that technically necessary? Nope.
Does it make a fleet of agents much easier to read at a glance? Absolutely.
That matters more than it sounds. Once you’re running specialized profiles for research, coding, outreach, operations, or anything else, the hard part isn’t creating another agent. The hard part is keeping the whole system legible. A distinct name, face, role, and conversation history gives your brain something concrete to grab onto.
Routines Might Be the Sleeper Feature
Bot Mode also adds a Routines pane for recurring tasks. Underneath, these are normal Hermes cron jobs namespaced to the bot that owns them.
That implementation choice is excellent.
If you tell one bot to summarize your inbox every morning, the routine lives beside that bot in the UI, but it still shows up in the normal Hermes cron system. The results land in that bot’s own chat history.
Nothing is being hidden inside some plugin-only scheduler. Bot Mode gives cron a better interface without creating a second automation stack you now have to manage.
This is exactly how plugins should work. Improve the surface. Reuse the primitives underneath.
Bot-to-Bot Messaging Is the Killer Feature
This is the part that pushes Bot Mode beyond a prettier profile switcher.
Every bot gets a persistent Agent Inbox conversation. Bots can send messages to other bots with clear attribution, and the receiving bot’s SOUL.md is taught the messaging protocol so it knows how to respond.
You can also use @mentions from a normal chat. Type something like @researcher check this claim and the active bot hands the request to that profile, waits for the answer, and brings the response back into the conversation.
Under the hood, these aren’t fake UI events. Bot-to-bot messages use real Hermes cross-profile chat handoffs. That means the specialist receiving the request brings its own memory, skills, model configuration, credentials, and personality to the job.
This is where the whole profile idea starts turning into an actual operating model. I don’t want one giant agent carrying every tool, every instruction, and every responsibility in one bloated context. I want a good default operator that can call the right specialist when it needs one.
Bot Mode makes that architecture visible.
The Architecture Is Cleaner Than the UI Suggests
The best technical decision here is that Nous didn’t patch Hermes core just to make this work.
The plugin uses the existing profile RPCs for listing, creating, describing, and configuring profiles. Avatar generation uses Hermes’ image generation RPC. Routines are standard cron jobs. Cross-profile chats handle bot messaging. Profile state remains under the normal ~/.hermes/profiles/<name>/ structure.
According to the official repo, there are no extra background daemons and no separate storage layer for the actual agent state.
That’s important because Bot Mode feels like a big new subsystem in the UI, but it’s mostly composition. Nous took things Hermes already knew how to do and snapped them together into a much better operator experience.
Where Hermes Bot Mode Still Feels Like a Beta
This is still a public beta, and there are a few places where that shows.
First, this is a desktop-side plugin. If your Hermes gateway lives on another machine, you install Bot Mode on the computer running Hermes Desktop, not on the remote gateway box. That distinction is going to trip people up.
Second, deleting profiles isn’t exposed in the Bot Mode UI. You still use hermes profile delete <name> from the CLI. I actually prefer that over making destructive profile deletion too easy, but it does mean the interface isn’t completely self-contained yet.
Third, bot-to-bot communication isn’t a live interrupt system. A receiving bot gets the message through its Agent Inbox invocation. Nous explicitly calls live interruption of an agent that’s already mid-conversation future work.
There’s also some version dependency. The roster itself can degrade gracefully on older gateways, but advanced profile editing and AI avatar generation need the newer profile and image RPCs available in recent Hermes builds.
None of these are deal breakers. They’re exactly the kind of edges I’d expect from a first public beta. But they’re worth knowing before you install it and assume the entire multi-agent experience has suddenly become point-and-click.
How to Install Hermes Bot Mode
The install is refreshingly simple. On the machine running the Hermes Desktop app, clone the official plugin into the desktop plugins directory:
git clone https://github.com/NousResearch/Hermes-Bot-Mode ~/.hermes/desktop-plugins/hermes-botsThen open the Hermes Desktop command palette with Ctrl+K or Cmd+K and run Reload desktop plugins. Restarting the app works too.
You should see a Bots tab appear beside Sessions, along with the Routines surface next to the conversation.
If you’re running a remote Hermes gateway, remember the important bit: install this where Desktop runs. Installing it only on the gateway machine won’t load the UI plugin on your laptop or workstation.
The official repository is available at NousResearch/Hermes-Bot-Mode on GitHub.
Who This Is For
If you only use one default Hermes profile and mostly treat it like a chatbot with tools, Bot Mode is cool, but it probably isn’t transformational yet.
If you’re already building a fleet, this is different.
Specialized profiles are one of the cleanest ways to keep agent context, permissions, skills, and responsibilities from turning into soup. Bot Mode gives that architecture an interface normal humans can actually navigate without mentally mapping profile names to terminal commands all day.
It also makes Hermes easier to explain. “Here are my agents, here’s what each one does, and they can talk to each other” lands a lot faster than a tour through profile directories, config scopes, cron namespaces, and cross-profile sessions.
Hermes Bot Mode Review Verdict
Day-one score: 4.5 out of 5.
I’m docking half a point because it’s still clearly beta software. Installation is manual, a few management actions still live in the CLI, and agent-to-agent communication isn’t yet the fully live shared-room experience this UI naturally makes you want next.
But the core idea is dead-on.
Hermes already had profiles. It already had isolated memory and skills. It already had cron. It already had cross-profile sessions. Bot Mode takes those pieces and turns them into something that feels coherent instead of merely powerful.
That’s why I think this matters beyond the cute avatars and pets. This is a glimpse of what Hermes Desktop looks like when multi-agent operation stops being an advanced configuration pattern and becomes the default way you interact with the system.
If you’re already running more than one Hermes profile, I’d install this beta immediately. Just go in knowing what it is today: a very smart UI layer over real Hermes primitives, not a replacement for them.
