agentos.guide › blog

Hermes vs Manus In Your Agent OS: Run Hermes As The Back Office And Manus As The Studio, With 5 Copy-Paste Prompts

By Julian Goldie · 8 November 2026 · agentos.guide

Hermes vs Manus inside an Agent OS isn't a fight for one seat: Hermes makes a strong back office that runs on your own machine and answers in your chat apps, and Manus makes a strong studio for research, websites and media.

Give each agent the jobs it was built for and neither has to fake the other's strengths.

Hermes Agent is open-source software from Nous Research under the MIT licence, and Manus is a hosted agent that works in a cloud sandbox or, through its desktop app, in a folder on your computer.

This guide gives you a quick start for both, five prompts to paste, and a weekly routine for a dog-grooming business.

If you'd like my team to look over your stack with you, book a free SEO strategy session.

Who does what in the Agent OS

Neither company publishes this split, so it's my reading of the Hermes docs and the Manus documentation.

No official page connects the two agents directly, so in this setup you carry work between them.

Quick start: Hermes as the back office

1. Install with curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash, or the PowerShell line on Windows.
2. Run hermes model and pick a provider, or point it at a local Ollama or LM Studio server.
3. Run hermes gateway setup, connect Telegram or WhatsApp, then hermes gateway install so it runs as a service.
4. Fill in an allowlist such as TELEGRAM_ALLOWED_USERS so only your team can give it orders.
5. Set approvals.mode: manual in ~/.hermes/config.yaml while you learn its habits.

Quick start: Manus as the studio

1. Sign in on the web and try one research task in the default cloud sandbox.
2. Install Manus Studio on an Apple Silicon Mac or a Windows PC if you want it working on local files.
3. In Studio, switch on Work locally, choose a small folder and set Permissions to Ask for approval.
4. Open Connectors and enable only the apps the first job needs.
5. Create a Project with a master instruction so every new task starts with your brand notes.

🔥 Want the Agent OS that ties both agents together?

Inside the AI Profit Boardroom you get step-by-step video trainings, the installable Agent OS, four coaching calls a week and 3,400+ members running Hermes beside hosted agents.

→ Join the AI Profit Boardroom

Settings to check before the first run

A few switches decide how safe the first week feels, so check them before you paste any prompt.

• Hermes allowlist: without one, anyone who finds the bot could message it, so set the platform's allowed-users variable first.
• Hermes approvals: the default smart mode lets an auxiliary model judge risk; switch to manual while you learn.
• Hermes terminal backend: the configuration page lists local, Docker, SSH, Modal, Daytona and Vercel Sandbox, and a container keeps experiments away from your own files.
• Manus permission mode: Approve for me is the default in Studio; pick Ask for approval for anything touching client folders.
• Manus folder choice: give each local task the smallest folder it needs, never your whole home folder.
• Manus connectors: enable only the apps a job needs, since each one gives Manus read and action access during tasks.

Prompt 1: the Hermes morning check-in

Paste this into your Hermes chat once and it sets up its own cron job.

Create a cron job called "Morning check-in" that runs at 07:30 Monday to Saturday.
Each run:
1. Read ~/business/bookings/today.csv and list today's appointments in time order.
2. Read ~/business/notes/ for anything dated today.
3. Flag gaps longer than an hour and any client marked "new".
Deliver the summary to this chat.
Draft only: never message a client.

The cron docs say Hermes manages jobs through its cronjob_manage tool, so asking in plain language is enough.

Prompt 2: lock down Hermes approvals

Open ~/.hermes/config.yaml and show me the approvals block.
Set approvals.mode to manual and keep cron_mode on deny.
Do not change anything else. Show me the diff before you save.

Manual mode asks you about every dangerous command, and cron_mode: deny blocks risky commands inside scheduled runs, per the security page.

Approvals cover commands, not messages, which is why every prompt here says "draft only".

Prompt 3: a Manus triggered task for booking emails

In Manus, open Automations, then Create, then Advanced automation, and paste this.

When a new email arrives in my connected Gmail with "booking" or "appointment" in the subject:
- Pull out the client name, pet name, service asked for and preferred day.
- Add one row to the Bookings spreadsheet in my connected Google Workspace.
- If any field is missing, flag it in the row instead of guessing.
Send me a short summary in Manus. Do not reply to the client.

The Automations docs ask you to review the timing, conditions and accounts Manus generates before you switch it on, and to use Test run with one email that should match and one that shouldn't.

Prompt 4: Wide Research on a supplier list

Use Wide Research on these 25 pet-grooming product suppliers: [paste list].
For each one, find: product range, whether they sell to salons, delivery regions and a contact page link.
Return one table, one row per supplier, and mark any field you could not confirm.

Wide Research hands every supplier to its own agent and then merges the results, which suits a list of this length.

Prompt 5: write one skill both agents can read

Hermes skills follow the agentskills.io standard, and Manus says it has adopted the same Agent Skills standard, with SKILL.md at the centre.

Write a skill at ~/.hermes/skills/booking-replies/SKILL.md.
It should hold our services, opening hours, groomer days and the tone for replies.
Add a rule: replies are drafts for a person to send.
Keep it under one page and show it to me before saving.

Give that prompt to Hermes, then upload the same folder to Manus from its Skills page and test it on one task in each.

Each product loads skills its own way, so check the output on both before you trust it.

📞 Want a second pair of eyes on how you split the work?

Book a free SEO strategy session and my team will go through which jobs belong on your own machine and which belong in the cloud.

→ Book a free call with my team

A weekly routine for a dog-grooming business

• Every morning: Hermes posts the check-in from Prompt 1 to the owner's Telegram.
• All day: the Manus automation from Prompt 3 logs new booking emails in the sheet.
• Monday: the owner runs Prompt 4 in Manus when restocking, then saves the table to the shared folder.
• Wednesday: Hermes drafts replies to any unanswered WhatsApp enquiries using the skill from Prompt 5, and the owner sends them.
• Friday: the owner asks Manus for a one-page social post plan for next week, and reviews it before anything is scheduled.

Every customer-facing step ends with a person pressing send.

The split works because each agent does what it is strongest at: Hermes sits in the owner's chat app all day, and Manus handles the bigger research and content jobs on demand.

If the business grows, a Manus Project keeps brand notes for every new task, and Hermes profiles keep separate setups side by side, for example one per team.

Common mistakes when running both

• Letting both agents watch the same inbox. Pick one agent per inbox, or you'll get two summaries and two sets of drafts for every email.
• Forgetting where each job runs. A Manus Studio task only runs while the computer is on and Studio is open, while a Hermes cron job needs its own machine awake.
• Trusting a community skill unread. Manus tells you to have it review a community skill's files before use, and the same habit protects Hermes.
• Skipping the test run. Run every new Manus automation and every new Hermes cron job once by hand before you leave it alone.

Most of these come down to one rule: each job should have a single owner in your Agent OS.

Handing work between the two

A shared folder is the simplest bridge: Manus Studio and Hermes can both work in a folder on the same computer.

Manus also offers a REST API for developers, but no official guide wires it into Hermes, so treat that as a custom build.

For a cloud-only comparison, see the Grok Bot vs Manus guide.

The ready-made Agent OS behind this routine is inside the AI Profit Boardroom.

FAQ

Which agent should sit at the centre of my Agent OS?

Hermes, because it lives in your chat apps and on your own machine. Send research, website and media jobs to Manus.

Can Hermes start Manus tasks automatically?

No official guide connects them. Manus has a REST API for developers, but wiring it into Hermes would be your own build.

Can I use the same skill in Hermes and Manus?

Both follow the Agent Skills format with a SKILL.md file, so one skill is a sensible starting point for both. Test it in each before relying on it.

Will these prompts send messages to my clients?

No. Each prompt says draft only or do not reply, and a person sends every customer-facing message.

Final word on Hermes vs Manus In Your Agent OS

Paste Prompt 1 tonight, build one Manus automation tomorrow, and book a free strategy session if you want help running both as one stack.

Give each agent its own lane and Hermes vs Manus stops being a choice you have to make.

Related guides on agentos.guide

→ Hermes Agent For Business: Start With One Job, Then Grow It With Five Copy-Paste Prompts In Your Agent OS

→ Hermes Agent For Accounting: 5 Prompts In Your Agent OS

→ Hermes Agent For Ecommerce: 5 Shop Prompts In Your Agent OS

→ Hermes Agent For Real Estate: 5 Prompts In Your Agent OS

📺 Video notes + links to the tools 👉

🎥 Learn how I make these videos 👉

🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉

← all guides on agentos.guide