Updated September 7, 2026: I originally reviewed Hermes Bot Mode when it was a brand-new public beta and a separate Desktop plugin. A lot has changed since then. Bot Mode now ships built into Hermes Desktop, Group Rooms are real, multi-machine fleets are first-class, and bot-to-bot messaging has grown into a much more serious coordination layer.
The funny part is that my original thesis survived.
Hermes profiles have been powerful for a while. They isolate memory, skills, credentials, models, sessions, and personality into separate agents. What Bot Mode did was make that architecture legible. Instead of profiles feeling like folders and configuration scopes, they became named agents you could actually talk to.
That was the promise in August.
A few weeks later, Hermes has pushed the idea much further.
Bot Mode is no longer just a nicer profile switcher. It is becoming the place where a fleet of Hermes agents starts behaving like an actual team.
What Hermes Bot Mode Is Now
The most important thing to understand has not changed: a Bot is still a real Hermes profile.
There is no fake second agent system sitting on top of Hermes. Each Bot keeps its own isolated configuration, memory, skills, credentials, model choice, sessions, and chat history. Bot Mode is the operator interface over those existing primitives.
That architecture is still one of the best decisions Nous Research made here.
The difference is that Bot Mode itself is now a built-in, default-on part of Hermes Desktop. There is no separate plugin repository to clone and no manual install step anymore. Open Desktop and the Bots surface is already there beside Sessions.
The current official documentation is here: Hermes Bot Mode documentation.
That alone fixes one of the biggest problems in my original review. Bot Mode no longer feels like an optional experiment bolted onto Desktop. It feels like part of the product.
The Bots Pane Has Become a Real Fleet View
The original Bot Mode roster already did something deceptively useful: it gave each profile a face, name, role, latest-message preview, and persistent chat.
The current version goes much further.
You can search the roster, see which Bots are active, hide agents you do not need in front of you, organize them into your own sections, and keep the same agent identity visible across Desktop connections.
That matters once your fleet grows.
One research agent, one coding agent, one operations agent, one outreach agent? Easy enough.
Add client-specific agents, machines in different locations, routines, temporary specialists, and a couple of agents with overlapping names, and the problem becomes less about creating agents and more about keeping the organization understandable.
Bot Mode is clearly being designed around that reality now.
I especially like the sections system because it gives you a second layer of organization without changing the underlying profile structure. You can group Bots into things like Clients, Development, Research, or whatever matches the way you actually work.
The UI is starting to look less like a profile browser and more like an operator console.
Creating a Bot Is Much More Complete
The fast path is still simple: name, title, description, done.
But the advanced creation flow is now substantial.
You can clone an existing profile or start fresh, pin a model and provider, write a custom `SOUL.md`, choose exactly which skills the Bot gets, enable specific toolsets, select MCP servers, and control how credentials are shared.
That is the part I care about most.
A specialist agent should not automatically inherit every capability in your stack just because it exists.
If I am creating a research Bot, I want research tools. If I am creating a coding Bot, I want the development surface. If I am creating something that touches production or client systems, I want to be deliberate about what it can reach.
Bot Mode now exposes much more of that capability design where normal users can actually see it.
There is also a Create on picker when Desktop knows about multiple Hermes backends. That means you can create a Bot directly on another registered machine without switching your entire Desktop session over to that machine first.
That is a huge deal for multi-machine Hermes setups.
Bot Identity Is No Longer Cosmetic
The avatar stuff is still fun. Blob faces, geometric faces, uploaded images, generated portraits, pixel pets. None of that makes the agent smarter.
It still matters.
Once you have a roster of specialists, visual identity helps your brain keep the organization straight. Names, roles, faces, and persistent conversations make each agent feel distinct enough that you stop thinking about profile directories and start thinking about responsibilities.
But the more important identity change is underneath the UI.
Every Bot now has a canonical persistent Bot Chat.
That conversation is treated as the long-term relationship with that agent. Hermes specifically protects it from accidentally turning into a disposable scratch session. If you try to use `/new` or `/reset` inside the canonical Bot Chat, Bot Mode redirects that into compaction instead of forking the relationship.
Fresh context, same agent, same conversation identity.
That is a small implementation detail with a big product implication.
Hermes is treating Bots more like persistent entities and less like temporary chat tabs.
Group Rooms Change the Whole Review
This is the biggest difference from the version I originally reviewed.
Group Rooms are real now.
You can put two to six Bots into a shared room and give them a task together. A message can trigger up to three serial rounds of agent turns. Mention a specific Bot and that agent gets priority. Mention nobody and the room can let the whole team weigh in.
Bots can pull another teammate into the discussion with `@name`, pass when they have nothing useful to contribute, and escalate an actual judgment call back to you with `@user`.
Hermes also puts hard caps on the loop so a room cannot sit there debating itself forever.
This is the feature that finally makes the original “these agents feel like a team” argument concrete.
Before Group Rooms, Bot Mode mostly made specialists easier to access and message.
Now you can create a research agent, a skeptical reviewer, and an implementation agent, put them in one room, and ask them to solve a problem together while each keeps its own memory, tools, model, and instructions.
That is much closer to the multi-agent setup I actually want.
I do not want one enormous agent carrying every skill, every credential, every instruction, and every responsibility in one bloated context.
I want specialists with clear jobs and a clean way to coordinate.
Group Rooms are the strongest expression of that idea Hermes has shipped so far.
Bot-to-Bot Messaging Got Much More Serious
Direct agent messaging has matured too.
Every canonical Bot Chat can use `message_agent` to send work to another Bot. The receiving agent gets clear attribution, runs inside its own canonical conversation, and the reply comes back as a background completion instead of pretending both agents are one giant shared session.
That separation is important.
The research Bot remains the research Bot. The coding Bot remains the coding Bot. Each brings its own model, memory, tools, skills, credentials, and standing instructions to the handoff.
You can also use normal `@mentions` from chats, and Hermes resolves those against the live Bot roster.
Cross-profile coordination is no longer an interesting trick. It is becoming part of the operating model.
There are still limits. Direct bot delivery is per invocation, not a live interrupt into an agent that is already mid-turn. But retries and failure handling are considerably more mature now, with typed reasons for things like quota failures, provider errors, missing configuration, offline runtimes, and timeouts.
That is the kind of boring infrastructure work that makes agent systems usable in the real world.
Multi-Machine Bot Mode Is the Sleeper Upgrade
This might be the most important change for people running Hermes seriously.
Desktop can now maintain a roster across multiple registered backends, including local runtimes, remote gateways, SSH-connected machines, Hermes Cloud instances, and other connections.
A Bot's memory, routines, sessions, and tools stay on the machine that owns that profile. Desktop does not pretend everything has been merged into one giant backend.
Instead, the roster gives you one operating surface across those machines.
You can create an agent on another machine, talk to it without switching the whole Desktop connection, put agents from different machines into the same broader workflow, and disambiguate duplicate names with device-aware handles.
That is a massive step forward from the original Bot Mode release.
For cross-machine direct messages, Desktop can act as the courier while it has both connections open.
For always-on gateway-to-gateway communication without Desktop in the middle, Hermes now also has `hermes peer`.
Peers let one Hermes gateway send work directly to another gateway's agent. Short messages can run as DMs, longer tasks can run asynchronously with a run ID, and Bots can learn about peer targets through the same messaging protocol used inside Bot Mode.
That gives Hermes two useful coordination paths instead of forcing everything through one surface.
Desktop is great when you are actively operating the fleet.
`hermes peer` is the more interesting answer when machines need to talk while Desktop is not acting as the control plane.
Routines Are Still One of My Favorite Parts
Routines were one of the sleeper features in the original review, and I still think that holds.
A recurring task belongs to the Bot responsible for doing it.
“Summarize my inbox every morning” should live beside the agent that owns email operations. A monitoring job should belong to the monitoring Bot. A content research routine should live with the research agent.
Underneath the UI, these are still normal Hermes cron jobs, namespaced to the Bot. They remain visible through the core cron system, and the results land back in that Bot's own history.
That is exactly how I want this implemented.
Bot Mode improves the operator experience without inventing a secret scheduler that only the Desktop plugin understands.
The surface gets nicer. The primitives stay reusable.
The Architecture Still Makes Sense
This was the strongest technical point in the first review, and it aged well.
Bot Mode did not replace Hermes profiles. It made them understandable.
Routines still use cron. Bot identities still map to profiles. Bot memory still lives with the owning profile. Skills and credentials stay scoped. Chats remain real Hermes sessions. CLI parity still exists.
You can talk to the same Bot from a shell with:
hermes -p <bot> chatYou can inspect its files under the normal profile directory. Its routines still appear in `hermes cron list`.
Turn Bot Mode off in Desktop and your underlying profiles, sessions, and cron jobs do not disappear.
That is a very good property for an agent UI to have.
Where Bot Mode Still Needs Work
The rough edges are different now.
The original review complained about manual installation, missing profile deletion, and the lack of a real shared-room experience.
Most of those complaints have been erased.
The current problems are more ambitious because the product is more ambitious.
Distributed group orchestration is still the obvious frontier.
Today, the full group-room orchestration log still has important Desktop-local behavior even though room metadata and recent history can sync through connected gateways. Nous is actively working through what it means for rooms to remain authoritative, synchronized, and alive when the Desktop client disappears.
There are also still edge cases around remote group rooms, cross-source routing, room UI parity, and keeping the custom group composer perfectly aligned with the normal Desktop composer.
You can see that work happening in the open repository. Two useful examples are the issue around exposing Bot Mode group rooms through the web dashboard and gateway and the design work around gateway-hosted group-room orchestration.
That is where the beta feeling lives now.
Not “how do I install this plugin?”
More like “what happens when six agents across several machines need to keep coordinating after the client that started the room disappears?”
That is a much better class of problem to have.
Do You Need Bot Mode?
If you use one Hermes profile as a general assistant and never plan to specialize it, Bot Mode is still optional.
The moment you start separating responsibilities, it becomes much more compelling.
A developer.
A researcher.
An operations agent.
A security reviewer.
A content agent.
A client-specific specialist.
Once those identities matter, Bot Mode gives you a way to manage them without mentally translating every task back into profile names, terminal commands, and config scopes.
And Group Rooms make the value proposition broader than it was in August.
You do not need a giant fleet to benefit anymore. Even three well-designed specialists can make a shared room useful.
Hermes Bot Mode Review Verdict
Updated score: 4.5 out of 5.
Yes, I am keeping the same score I gave the original beta.
But I am docking the half point for completely different reasons now.
The original Bot Mode lost points because it still felt like an early plugin: manual installation, incomplete management controls, and agent coordination that clearly wanted a shared-room layer it did not have yet.
Those problems are mostly gone.
The current version is dramatically more capable. It is built into Desktop, agent management is much better, canonical Bot Chats give agents durable identity, Group Rooms are real, direct messaging is mature, and multi-machine operation is becoming a first-class part of the design.
Now the missing half point is about distributed reliability and control-plane maturity.
Group Rooms are powerful, but the hardest multi-machine and Desktop-off orchestration problems are still being worked through. The product has moved from “cool multi-agent UI” into much harder territory, and some of those edges are still visible.
That does not weaken the core idea.
If anything, the last few weeks proved the original thesis.
Hermes already had the pieces for specialized agents. Bot Mode turned those pieces into recognizable individuals.
Group Rooms, persistent Bot Chats, routines, multi-machine rosters, and direct agent messaging are now turning those individuals into something closer to an actual team.
That is the reason I would use Bot Mode today.
Not because the avatars are fun.
Not because multi-agent sounds impressive on a feature list.
Because once your AI system grows beyond one general-purpose assistant, you need a way to make responsibilities, identity, coordination, and boundaries visible.
Hermes Bot Mode is getting very good at exactly that.
For the current release notes, see Hermes Agent v0.21.0.
