Your Hermes Agent will run bash, hold your API keys, and execute code an LLM wrote that no human reviewed. Running that in a shared Docker container — where every tenant sits on one host kernel — is the wrong default. Here's the alternative: the Nous Research Hermes Agent in its own Firecracker microVM, real hardware isolation, your model key never leaving the box, provisioned automatically the moment you pay — the microVM boots in 124ms. This guide covers what Hermes Agent is, why a real microVM beats a Docker container or a credits-bundled box, and exactly how to get yours running.
What the Hermes Agent actually is
Hermes Agent is Nous Research's long-running, self-improving agent. It is not a chat window. It creates skills from what it does, keeps a memory that compounds over time, and connects to Telegram, Discord, Slack, WhatsApp, Signal, and email through a single gateway. It runs tools, browses, executes code, and schedules its own recurring work.
The important part for hosting: it is model-agnostic and bring-your-own-key. You point it at OpenRouter, the Nous Portal, OpenAI, NVIDIA NIM, or your own endpoint. Nobody sits between your agent and your model provider.
Who runs it on a server instead of a laptop? People who want the agent awake when they are not — answering messages overnight, running scheduled jobs, holding a memory that does not reset when the laptop sleeps.
Want the official Electron GUI instead of the web dashboard? That is a different surface on the same ~/.hermes — see Hermes Agent Desktop and the Hermes Desktop landing.
Why a real microVM, not a Docker container
Most "one-click Hermes" offers hand you a Docker container on a shared host, or a box with model credits bundled in. Both quietly cost you something.
A container shares the host kernel. There are over 200 known weaknesses in the container stack, and open research on whether a capable model can talk its way out of a sandbox. That is the wrong boundary for a process holding your credentials and running code an LLM returned. jurniti gives every agent its own Firecracker microVM with its own kernel and a KVM hardware boundary — the same isolation primitive AWS uses under Lambda, one per tenant, booting in about 124ms. Not Docker theatre.
Bundled credits — and even most "BYOK" — put a middleman on your keys. Hosts that "pre-install AI credits" charge their margin on every token. And the common BYOK gateway still stores your key and proxies your model traffic through it. jurniti's BYOK is different by architecture: you paste the key inside your own VM, model traffic leaves that VM straight to the provider, and jurniti is never in the path. No proxy, no token markup, you own the provider relationship.
The third thing you get is a reversible decision: a 30-day money-back guarantee, with your persistent volume retained for 7 days after cancellation in case you change your mind.
Deploy Hermes Agent in 3 steps
Setting this up by hand on a raw VPS is an afternoon of work — non-root user, installer, provider config, a systemd unit, a reverse proxy for TLS — and then it's a standing job: kernel patches, TLS renewals, and the pull-rebuild-restart cycle every time upstream ships. jurniti runs the box ops so you don't, and you still own the agent version. Here it is in three steps, provisioned automatically the moment you pay — the microVM boots in 124ms and the agent installs on top.
1. Pick a plan and pay
Choose Starter, Pro, or Max and check out. The moment Stripe confirms, the provision chain starts automatically. No human touches your box.
2. Your microVM boots and Hermes pre-stages
A fresh Ubuntu microVM boots and the upstream Hermes installer runs. You get an in-browser terminal in the dashboard, so there is nothing to SSH into. On Pro and up, the Hermes Dashboard also gets its own TLS subdomain at https://<you>.jurniti.com.
Set your model key once. The simplest path is the interactive picker, which walks you through any provider:
hermes model
Or set an OpenRouter key directly:
hermes config set OPENROUTER_API_KEY sk-or-v1-...
3. Talk to your agent
Open the dashboard, connect a channel (Telegram is the fastest), and start giving it work. Your skills, memory, and config live on the persistent home volume — they survive every restart and upgrade.
What people build with a hosted Hermes Agent
- An always-on inbox triager. Connect email + Telegram; it sorts, drafts replies, and briefs you each morning.
- A research analyst. It compares tools or vendors on demand and keeps the comparisons in memory so the next question is cheaper.
- A multi-platform support bot. One agent answering Discord and WhatsApp, escalating only what needs a human.
- A scheduled operator. Built-in cron runs recurring jobs — daily reports, scrapes, syncs — without you remembering to trigger them.
Because skills persist, the agent you have in month three is more useful than the one you started with. You can also snapshot a configured agent into a template and fork it into a second VM.
jurniti vs a DIY VPS vs Docker-template hosting
| jurniti | DIY VPS | Docker-template host | |
|---|---|---|---|
| Isolation | Firecracker microVM (KVM) | Shared kernel | Shared kernel (container) |
| Model keys | Yours, never leave the VM | Yours | Often bundled + marked up |
| Setup | Automatic · ~3 min to live | An afternoon | One click |
| Persistent home | Yes, survives restarts | You configure it | Varies |
| Refund | 30-day money-back | None | Varies |
| You own the upstream | Yes — real Nous installer | Yes | Sometimes a fork |
Pricing
Starter is $25/mo ($250/year), Pro $49/mo ($490/year), Max $99/mo ($990/year) — annual billing is two months free. Every plan includes the 30-day money-back guarantee. Model spend is separate — you pay your provider directly, no markup.
Running a different agent? See the guides for OpenClaw, Pi, and Claude Code, jump straight to managed OpenClaw hosting, or compare plans on the pricing page.
