Nous Research's Hermes Desktop Turns One AI Into a Full Agent Team

Nous Research just shipped Bot Mode for its open-source Hermes Desktop app, and it turns the single-assistant chat window into something closer to a Slack workspace of AI teammates. Bot Mode replaces the single-agent session list with a roster of named bots. Each bot is a real Hermes profile, with its own chat, memory, skills, and pinned model.

The pitch is simple: instead of spinning up a fresh session and re-explaining context every time, you build a specialist once and keep it around. Users can build a specialist agent once and return to that same bot over time.

One profile, one bot, one lane

Under the hood, Bot Mode is a thin UI over a primitive Hermes already had: profiles. Each bot operates as an isolated Hermes profile stored on disk at ~/.hermes/profiles/name/. This keeps memory, configuration, skills, credentials, and chat histories separated without adding a new storage layer.

What you actually configure per bot:

• A name, title, description, and avatar (geometric, uploaded, or AI-generated)
• A pinned model, so a research bot can run a big frontier model while an inbox bot runs a small local one
• Per-skill and per-toolset enablement
• A custom SOUL.md persona file that defines behavior and protocols
• Cron-backed routines scoped to that bot

Users can customise each bot's avatar, model pin, enabled skills, custom SOUL.md persona file, and routine schedules. Routines are just namespaced Hermes cron jobs, so a scheduled "summarize my inbox every morning" runs inside the bot that owns the relevant context and lands in its own chat history.

The real trick: bots talking to bots

The headline feature is agent-to-agent messaging, and the implementation is refreshingly unglamorous. Bots communicate with one another using a persistent Agent Inbox. Handoffs are executed via CLI commands (e.g. hermes -p bot chat -c "Agent Inbox"), triggered when a user or bot uses an @mention.

In practice, you type @researcher have a look at this in any chat, the active bot hands the message off, waits for the reply, and reports back. Messages arrive with attribution like [Message from agent 'researcher'], and each bot's SOUL.md teaches it how to reply to inbox traffic. No new daemon, no new database, no orchestration framework layered on top.

What is now possible

The practical shape of this is a small AI team you actually reuse:

• A research bot with web search enabled that supplies cited sources to a writing bot
• A developer bot handling technical questions delegated from a planning bot
• A scheduled briefing bot that runs a cron every morning and pings you
• A local, cheap model handling triage that escalates to a bigger model via mention

A research bot can supply sources to a writing bot, a developer bot can handle technical questions from a planning bot, and scheduled routines can run inside the agent that owns the relevant context. Because each bot can pin its own model, you can route by cost and capability inside a single desktop app instead of writing your own router.

Price, availability, limits

Yes , today, on a desktop, at no license cost. Bot Mode shipped first as a one-day public beta plugin. It is now bundled and default-on inside Hermes Desktop, under Settings → Plugins. Both Hermes Agent and the plugin are MIT licensed. Hermes Desktop itself is free and runs on macOS, Windows, and Linux, with a separate paid Nous Portal tier if you want managed inference credits.

A few sharp edges are worth knowing before you build on it. Solo builders, startups, and small-to-mid engineering teams can adopt it immediately. Enterprises should treat it as a workstation tool, not managed infrastructure. Bot-to-bot delivery is also per-invocation, meaning the receiving bot sees a message the next time it runs; you cannot live-interrupt a bot mid-conversation yet. And profile deletion is intentionally kept out of the UI, so you drop to hermes profile delete name in the terminal for cleanup.

Why this matters

Multi-agent frameworks are having a moment. Arriving just days after xAI's Grok Bot sensation, the release places Hermes in the same emerging category of multi-agent systems. The difference is that Hermes puts the workspace on your machine, under an MIT license, with no core patches required to extend it.

The design choice worth stealing is that a "bot" is not a new abstraction. It is the profile primitive Hermes already had, wrapped in a roster UI and connected by a CLI handoff. Bot Mode now ships built into Hermes Desktop. As of hermes-agent #87886, Bot Mode is a bundled, default-on Desktop plugin (toggle in Settings → Plugins) with the multi-source roster included. Development continues in-tree at apps/desktop/src/plugins/hermes-bots/. For anyone tired of rebuilding agent scaffolding every project, that is the interesting part.