What is Bot Mode for Hermes? Named bots, explained
Hermes Desktop turns agent profiles into a roster of named Bots, each with its own role, model, memory, and skills, messaging each other by @mention. How it works and what it costs to run.
Bot Mode is a Hermes Desktop feature that turns your agent profiles into a roster of named Bots, each with its own role, model, memory, skills, and avatar. Bots message each other, hand work off by @mention, deliberate in shared rooms of two to six, and run scheduled routines of their own. Underneath, a Bot is a Hermes profile (isolated config, memory, and history under ~/.hermes/profiles/<name>/), so the feature is a team-shaped surface over machinery Hermes already had. It ships built into the desktop app, on by default since mid-August 2026, MIT-licensed like the rest of Hermes.
What a Bot actually is
The docs put it plainly: Bot Mode “turns your Hermes profiles into a roster of named Bots,” and a Bot “is a Hermes profile — isolated config, memory, skills, credentials, and chat history.” The build-a-specialist pitch follows from that isolation. Each Bot carries:
A role and persona: what the Bot is for (“level designer,” “QA,” “narrative writer”), shaped through its own SOUL.md. Its own model: “any provider/model pair Hermes knows about works, and different Bots can run on different models side by side,” so a cheap fast model can do triage while a frontier model writes. Hermes’s any-model posture, local models included, is the difference from the one-provider bot rosters elsewhere. Its own memory: persistent, in its profile directory, not shared with its teammates. Its own skills: enabled per-skill, per-toolset, and per-MCP-server, so the accounting Bot never holds the deploy keys. An avatar: blob faces, shapes, an upload, or a generated portrait.
How Bots work together
@mentions route work. Type @researcher have a look at this in any chat and the mention resolves against the live roster; the active Bot is told exactly who you mean. Bots reach each other the same way, through a message_agent tool that delivers fire-and-forget messages, with attribution, into the teammate’s own chat.
Rooms bound the deliberation. Shared collaboration rooms hold two to six Bots, and a message into the room triggers up to three serial rounds of member turns, with hard caps (ten messages per send, three rounds) so a conversation between machines cannot run away with your model spend.
Routines are cron underneath. A Bot’s scheduled work is “plain Hermes cron jobs namespaced [bot:<name>] <routine>”. They show up in hermes cron list like any other job, which is good news for debugging: nothing about a Bot’s schedule is a new subsystem.
Creating one, and where it lives
The New Agent dialog needs a name, title, and description; advanced options add cloning an existing Bot, pinning the model, editing SOUL.md, and skill selection. On a desktop connected to several machines, a “Create on” picker chooses which machine hosts the new Bot, and a name that exists on several machines gets an @name-device handle. State lives on the host machine, in that machine’s ~/.hermes/profiles/. That is worth knowing before you build a specialist on a laptop that sleeps.
Availability, and two caveats
Bot Mode ships built into the Hermes desktop app and is on by default: no install, toggled under Settings → Plugins → Bots without a restart. It arrived bundled in mid-August 2026 after a brief public run as a standalone plugin, and carries the MIT license.
Two things to hold onto. First, every Bot is a model consumer: a six-Bot room running three rounds is up to eighteen model calls from one message, so roster size and room membership are budget decisions, not just organization. Second, the docs are quiet on scaling limits (maximum roster size, concurrent routine execution), so treat a large roster as unmapped territory. For orchestration with harder edges (a task board, capped concurrency, explicit spend limits), Hermes kanban is the sibling feature built for exactly that, and the two compose: kanban for the work queue, Bots for who does the work.