이미지: X — 모델·오픈소스 영상 갈무리
Summary
- Nous Research unveiled 'Bot Mode' for its agent app Hermes Desktop on August 17, 2026
- Existing agent profiles are now split into multiple named bots, each with its own role, model, memory, skills, and profile picture
- The company said bots can run on any model and can communicate with one another
- 발표 주체
- Nous Research
- 발표일
- 2026년 8월 17일
- 기능명
- Bot Mode (허미스 데스크톱)
- 봇 단위 구성 요소
- 역할·모델·기억(memory)·스킬·프로필 사진
- 회사가 밝힌 특징
- 봇이 임의의 모델 사용 가능, 봇 간 상호 통신 가능
Not profiles, but 'named colleagues'
The old approach — running a single AI assistant by swapping its settings for each task — is changing. On August 17, 2026, Nous Research unveiled Bot Mode for its agent app Hermes Desktop. What used to be managed as multiple "agent profiles" — bundles of settings — is now split into multiple named Bots.
Each bot has its own role and its own model. On top of that, each gets its own memory, skills, and even its own profile picture. The company said bots can run on any model and can talk to one another. The announcement summed it up in one line: "Build a specialized bot once, keep using it."
Why split the 'profile'
An AI agent doesn't just answer questions — it does the work for you. It searches, creates files, even makes reservations. Add skills — bundles of instructions that teach the AI the steps for a specific task — and it can remember something like "our company's report template."
The problem shows up as these settings pile up. Load a single agent with code-review instructions, marketing-copy instructions, and a research tool list all at once, and they interfere with each other. Swapping settings for every task is also a hassle. Bot Mode breaks this pile apart into human-like, named units. A research bot only has the research bot's memory; a document-organizing bot only has the document-organizing bot's skills. The fact that each bot can be attached to a different model also matters in practice — it lets you assign a heavy reasoning model to a planning bot while a cheap, fast model handles a repetitive-task bot, all within the app.
This direction isn't unique to Hermes. On August 11, xAI released the Grok Bot agent as an early beta, and a day later a solo operator showed off a screen running actual work across nine Grok bots split by task. On August 10, Microsoft published a blog explaining how to build no-code agents in Microsoft 365 Copilot's 'Agents' panel. From "one all-powerful assistant" to "multiple role-split bots" — the industry's center of gravity has been tilting the same way over the past few weeks.
So what actually changes
On August 10, Hermes had already merged its 12 browser-control tools into a single tool built on browser_use CLI 3.0, saying its internal tests showed a 48-66% cut in token usage. Having slimmed down a single bot's toolset, it's now scaling that lighter bot out into many.
For users, what changes is accumulation. A well-built specialized bot no longer gets deleted — it keeps getting called on, with its conversation history and skills persisting inside it. If bots really do communicate with each other as described, you could build a pipeline right inside the app, like a research bot handing off gathered material to a drafting bot. Multi-agent setups once confined to developer frameworks have now come down to the level of profile pictures in a desktop app.
