The Hermes Agent backend is the setting that decides where your agent runs its commands, and picking the right one is the difference between a safe business tool and an agent with the keys to your laptop.
Hermes Agent is the free, open-source AI agent from Nous Research, and its docs use the word "backend" for six different layers.
The first layer is the terminal backend, which is where shell commands run, and the docs list local, Docker, SSH, Modal, Daytona, Vercel Sandbox and Singularity.
The other layers are the model provider, the web search provider, the browser provider, the memory provider, and Hermes itself acting as the backend for another app.
I'll explain each one in plain English first, and then I'll show you which choice fits a solo owner, an agency and a client-facing bot.
I took every fact here from the official Hermes docs and the read-only help output on my own install, so nothing below is a guess.
What The Hermes Agent Backend Actually Is
Think of Hermes as a worker and the backends as the rooms and suppliers that worker uses.
One backend is the room where the worker does the hands-on jobs.
Another backend is the brain the worker rents.
The others are the search service, the browser, the filing cabinet and the front desk.
| Backend type | Plain-English job | Options the docs list | How you set it |
|---|---|---|---|
| Terminal | It is the room where commands run. | The options are local, docker, ssh, modal, daytona, vercel_sandbox and singularity. | You run hermes config set terminal.backend <name>. |
| Model | It is the brain that does the thinking. | The options include Nous Portal, OpenRouter, Anthropic, OpenAI, Bedrock, Ollama, LM Studio and custom endpoints. | You run hermes model. |
| Web | It is the search and page-reading service. | The options include Firecrawl, SearXNG, Brave, DuckDuckGo, Exa, Parallel, Tavily and Perplexity. | You run hermes tools. |
| Browser | It is the browser the agent clicks around in. | The options include Browser Use, Browserbase, Firecrawl, Camofox, Lightpanda and local Chromium. | You run hermes tools. |
| Memory | It is the long-term filing cabinet. | The options include Honcho, OpenViking, Mem0, Holographic, RetainDB, ByteRover and Supermemory. | You run hermes memory setup. |
| Hermes as a backend | It is the front desk other apps talk to. | The options are the API server, ACP, MCP serve and hermes serve. |
You run the matching command. |
Hermes itself is free under the MIT licence, so the only costs come from the services you plug into those slots.
🔥 Want my exact Hermes Agent Backend For Business setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
The Terminal Backend In Plain English
When Hermes installs a package, edits a file or runs a script, that command has to run somewhere.
The terminal backend is that somewhere.
| Backend | Where the work happens | Who it suits |
|---|---|---|
| local | The work happens on your own computer. | It suits one person working on their own machine. |
| docker | The work happens inside a sealed container on your computer or server. | It suits anyone who lets other people message the agent. |
| ssh | The work happens on a remote server. | It suits a team that wants the agent off everyone's laptops. |
| modal | The work happens in a Modal cloud sandbox. | It suits short bursts of cloud compute. |
| daytona | The work happens in a Daytona cloud workspace. | It suits a managed cloud dev environment. |
| vercel_sandbox | The work happens in a Vercel Sandbox microVM. | It suits teams already on Vercel. |
| singularity | The work happens in an Apptainer container. | It suits research clusters where Docker isn't allowed. |
The default is local, and the docs call it the safest default when you're getting started because there is nothing to set up.
The docs also say that on local, the agent has the same filesystem access as your user account.
That sentence is the whole reason the other six exist.
Use Case 1: The Solo Owner On One Laptop
If it's only you, and you're the only person who talks to the agent, local is fine.
Your agent can use the logins you already have, such as git and npm, because commands run with your normal home folder.
Hermes still protects you here, because on local it checks commands against a list of dangerous patterns and asks before it runs a match.
For the brain, the docs recommend Nous Portal, where one login covers 300+ models and a Tool Gateway for web search, images, speech and the browser.
If you'd rather pay nothing for the model, you can point Hermes at Ollama on your own machine with the address http://localhost:11434/v1.
The docs warn that Ollama uses a small context window by default, so you set context_length in the config.
My full starter stack is in the best Hermes Agent setup.
Use Case 2: The Agency With A Team Bot
The moment staff can message the agent from Slack, Telegram or Discord, the risk changes.
Somebody else's message can now trigger commands on the machine that runs Hermes.
The docs give a direct tip for this, which is to use docker, modal, daytona or vercel_sandbox for production gateway deployments.
Docker is the cheapest of those, because it runs on a server you already have.
hermes config set terminal.backend docker
Hermes then runs every command inside a container with all Linux capabilities dropped, no privilege escalation and a limit of 256 processes.
It keeps one container alive, so anything the agent installs is still there for the next job.
If you want each conversation fully separated, you set container_persistent: false, and every chat gets a fresh container.
If the agent never needs the internet for its commands, you set docker_network: false, and the container is cut off from the network.
The other route for an agency is SSH, which sends the commands to a remote server instead.
The Hermes tools page recommends SSH for security because the agent can't modify its own code from a different machine.
I've compared servers for that in the best Hermes Agent VPS.
🔥 Want my Hermes setup for a real business? Inside the AI Profit Boardroom, I show how I run agents in my own agency, and you can ask me questions on four coaching calls a week alongside 3,400+ members. → Get access here
Use Case 3: A Chat App For Your Team Or Clients
This is where Hermes becomes the backend itself.
The API server turns your agent into an OpenAI-compatible endpoint, which means any chat app that speaks that format can sit on top of it.
The docs name Open WebUI, LobeChat, LibreChat, NextChat and ChatBox as examples.
You add two lines to ~/.hermes/.env and start the gateway.
API_SERVER_ENABLED=true
API_SERVER_KEY=change-me-local-dev
hermes gateway
The server then listens on http://127.0.0.1:8642, and your chat app connects to http://localhost:8642/v1.
The person typing gets more than a chatbot, because the agent answers with its terminal, files, web search, memory and skills.
That is exactly why this use case needs a sandboxed terminal backend behind it.
There are three other ways to put Hermes behind something else.
| Mode | Command | Business use |
|---|---|---|
| ACP server | hermes acp |
Your developers use Hermes inside VS Code, Zed or JetBrains. |
| MCP server | hermes mcp serve |
Another agent, such as Claude Code, reads and sends your team's messages through Hermes. |
| Subscription proxy | hermes proxy start |
Another app uses your model subscription without any agent tools. |
The docs describe the split clearly: use the API server when you want the agent as a backend, and use the proxy when you only want the model.
Use Case 4: Research And Content Work
Content and research jobs lean on two more backends, which are web and browser.
The web backend powers web_search and web_extract.
Firecrawl is the default, and a new install works with no key because Hermes rotates across the free tiers of Exa, Parallel, Firecrawl and Keenable.
That keyless tier is a last resort, so a business that depends on research should add a keyed provider for steady service.
Free options exist too, because SearXNG is free when you host it yourself and DuckDuckGo needs no key.
The browser backend is a separate choice for jobs that need clicking, logging in or filling forms.
The docs list Browser Use, Browserbase, Firecrawl, Camofox, Lightpanda, your own Chrome over CDP and a local Chromium.
I explain that toolset in the Hermes Agent browser guide.
Use Case 5: An Agent That Remembers Your Clients
Hermes has built-in memory in two files, MEMORY.md and USER.md, and that memory is always on.
A memory backend is an extra provider you add on top, and only one can be active at a time.
You set it up with hermes memory setup.
The docs list Honcho, OpenViking, Mem0, Holographic, RetainDB, ByteRover and Supermemory, and their comparison table shows a mix of free, paid, local and cloud options.
My ranked picks are in the best Hermes Agent memory.
Choosing The Model Backend For A Business
The model backend is where most of your running cost sits.
| Provider backend | How you connect it | Why a business picks it |
|---|---|---|
| Nous Portal | You run hermes setup --portal. |
One subscription covers models and tools. |
| OpenRouter | You add OPENROUTER_API_KEY to ~/.hermes/.env. |
One key reaches many models. |
| Anthropic | You pick it in hermes model. |
You already use Claude. |
| OpenAI API | You add OPENAI_API_KEY. |
You already have an OpenAI account. |
| AWS Bedrock | You pick "AWS Bedrock" in hermes model. |
Your company already buys through AWS. |
| Ollama or LM Studio | You pick the local option in hermes model. |
Your data must stay on your own hardware. |
| Custom endpoint | You enter a base URL and model name. | You run your own OpenAI-compatible server. |
You add a provider with hermes model in the terminal, and you switch between providers you've already added with /model inside a chat.
You can also add a fallback chain with hermes fallback, so the agent moves to a second provider if the first one fails.
If your company is on AWS, read Hermes Agent on Bedrock for the full steps.
Security, Speed And Cost Side By Side
| Question | local | docker | ssh | Cloud sandboxes |
|---|---|---|---|---|
| Is the host protected? | No, because commands run on the host. | Yes, because a container is the boundary. | Yes, because the work is on another machine. | Yes, because the work is in the provider's sandbox. |
| Does Hermes ask before dangerous commands? | Yes, it does. | No, the check is skipped. | Yes, it does. | No, the check is skipped. |
| Is there setup? | There is none. | You need Docker running. | You need a server and two settings. | You need an account and API keys. |
| What does it cost? | It is free. | It is free on your own machine. | You pay for the server. | You pay the provider's own pricing. |
On speed, local has no sandbox to start, Docker reuses one running container, and SSH reuses one connection.
Container backends start with 1 CPU core and 5 GB of memory by default, and you can raise both.
Cloud sandboxes save your files between sessions, and the docs say they don't save running processes.
One warning applies to Docker, which is that any secret you forward into the container can be read by code running inside it.
🔥 Ready to put agents to work in your business? The AI Profit Boardroom has my Hermes training, my agent setups and four coaching calls a week, so you can copy what works instead of guessing. → Join the AI Profit Boardroom
My Simple Decision Rule
- I'd stay on local if I'm the only person who talks to the agent.
- I'd move to Docker the day a second person can message it.
- I'd use SSH when I want the agent on a server that never sleeps.
- I'd look at a cloud sandbox only when I don't want to run a server at all.
- I'd switch on the API server when the team wants a chat app instead of a terminal.
Once the backend is settled, Hermes Agent background tasks shows how to keep the agent working after you log off.
Related Reading
- The best Hermes Agent tools covers the toolsets that run on these backends.
- Herald OS has a settings page where you choose where agents run tasks, including Docker.
- FlyHermes is a hosted route if you don't want to manage a backend yourself.
Also On Our Network
- 🌐 The exact commands to switch every Hermes backend
- 🌐 All seven Hermes terminal backends scored
- 🌐 Where Hermes backends sit inside an Agent OS
- 🌐 Which model to run behind your Hermes backend
FAQ: Hermes Agent Backend
What is the Hermes Agent backend?
It most often means the terminal backend, which decides where the agent runs shell commands.
The same word also covers the model provider, the web provider, the browser provider and the memory provider.
Which backend should a business use?
A solo owner can stay on local, and a business that lets other people message the agent should use a sandbox such as Docker.
Does the backend cost money?
Hermes is free, and local, Docker, SSH and Singularity only cost you the hardware you already have.
Modal, Daytona and Vercel Sandbox are paid services, and your model provider bills separately.
Can my team use Hermes through a chat app?
Yes, because the API server gives any OpenAI-compatible chat app a full agent on port 8642.
How do I switch the backend?
You run hermes config set terminal.backend with the backend name for the terminal, and you run hermes model for the provider.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I run Goldie Agency, a 7-figure SEO and link building agency based in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the book "Link Building Mastery".
- I run GoldieBench, where I score AI models on real one-shot builds.
→ Get my best AI training inside the AI Profit Boardroom
Match the room to the risk, match the brain to the budget, and you'll have the right Hermes Agent backend for your business.
