It stops when you close the lid
The assistant you pay for lives in a tab. Shut the laptop and the research job dies, the inbox goes untriaged, the scheduled thing never runs. You're paying for a capability that's only awake when you are.
HermesBox is a small always-on machine that runs your assistant on your desk. Plug in the Claude, GPT or Gemini key you already pay for — it uses your plan, your provider bills you, we take nothing. Local models for the work you'd rather not send anywhere.
The models are already good. The keys are already in your password manager. What's missing is a machine that stays awake and does something with them.
The assistant you pay for lives in a tab. Shut the laptop and the research job dies, the inbox goes untriaged, the scheduled thing never runs. You're paying for a capability that's only awake when you are.
A Claude plan. A GPT plan. Then some agent service on top that resells you the same tokens at a markup, in its own credits, on its own servers — for the privilege of running them while you sleep.
Contracts, patient notes, unreleased code. So you rewrite the real question into a vague one, get a vague answer back, and quietly do that part by hand instead.
This is the part people underestimate. HermesBox is not a small model pretending to be a big one — it's the machine your agent lives on, pointed at whichever provider you already have a relationship with.
Your API key or Claude plan. Billed by Anthropic, not us.
Drop in the key you already use for everything else.
Long-context work, on the account you already have.
One key, the whole catalogue, switchable per task.
Runs on the box itself. No key, no bill, nothing leaves.
Anything else with an OpenAI-compatible endpoint drops in the same way.
| ✓ | anthropic | claude | your key · your plan |
| ✓ | openai | gpt | your key · your plan |
| ✓ | gemini | your key · your plan | |
| ✓ | openrouter | 300+ models | your key · your plan |
| ✓ | ollama | local | on this box · no key |
Most things that sit between you and a model take a cut, or resell you capacity at a markup, or make you buy credits in their currency. This doesn't. Your key goes in, the box calls the provider directly, and the invoice arrives from Anthropic or OpenAI exactly as it did last month.
We sold you a computer once. That's the entire commercial relationship.
Filling a PDF properly. Cleaning a spreadsheet the way an accountant would. Reviewing a diff. Each one is a small folder of instructions your agent reads when the job comes up — so it stops guessing at the shape of the work and just does it the known-good way.
They run on whichever provider you pointed the box at — a skill that reads a contract can run against the local model and never leave the room, and the same skill can run on Claude tomorrow because you changed your mind, not because we changed our terms.
pdf/
├── SKILL.md ← what to do, in English
├── reference.md ← the fiddly bits
└── scripts/ ← optional helpers$ hermes skills add pdfOr just ask it in Telegram — same thing, fewer keystrokes
Buy the box, plug in the keys you already have, and never pay us again. If it isn't what you wanted, send it back inside thirty days for every cent — no form, no phone call.
In stockActual unit · not a renderYour AI agent, on hardware you own.
One-time payment. VAT and shipping included — no subscription, ever.
Free delivery in 2–5 working days — dispatched from Bulgaria to the EU, UK, Switzerland and Norway.
In stock. Dispatched from Bulgaria within 2–5 working days to the EU, UK, Switzerland and Norway. Shipping and EU VAT are included — the number above is the number you pay. Model usage is billed by your own provider on your own key; we add nothing to it and there is no subscription to us.
As a consumer you have a statutory right to withdraw within 14 days of delivery; we extend that to 30. A 2-year legal guarantee of conformity applies on top. Terms of sale · Privacy
A Claude plan. A GPT plan. And then a third invoice, every month, from a service whose entire contribution was keeping a process running while we slept — using the keys we'd already paid for.
That third invoice is a computer. So we made the computer, sold it once, and stayed out of the token business entirely.
We build the same box under a second name. clawbox.com sells the hardware on its own, for people who want the machine without the agent build on top. Same company, same workshop, same thermals — so if you have read one of these pages, you have read most of the other.
The cleanest audit there is: buy the box, run a week of work through it, then open your Anthropic or OpenAI billing page. What you see there is what you paid. Nothing routes through us, so there's nothing for us to skim.
Ollama, llama.cpp, open weights, an OpenAI-compatible endpoint, your own API keys. Every piece is something you could assemble yourself — we just did the assembling, the thermals and the enclosure.
Thirty days from the day it lands. Which means the most you can lose by being wrong about us is the postage back.
No Docker compose file, no CUDA versions, no quantization rabbit hole. The hard part is already done and soldered down.
Power and Ethernet. It boots, finds itself on your network, and answers to hermes.local. No account to create, no setup wizard, no waiting list.
Anthropic, OpenAI, Google, OpenRouter — whichever you're already paying for. The box uses your key and your plan; your provider bills you exactly what they billed you yesterday. We never see a token.
Or the web UI, or the OpenAI-compatible endpoint your existing tools already speak. It's awake at 3am with your laptop shut, because it isn't your laptop.
Same box, both routes — cloud when it helps, local when it matters
A Jetson Orin Nano Super with the models already on the drive. It draws less than the lamp next to it and does nothing else with its life.

Eight gigabytes is the honest constraint here: it runs 7–8B models at four-bit comfortably, and it does not run a 70B. If your work needs a frontier model, this is the wrong machine and we'd rather you knew that now than after it arrived.
Comparison is against a hosted agent service, not against the model providers themselves — you keep those. HermesBox is where they run, not a replacement for them.
No account to sign into. No terms that quietly change. No model swapped underneath you on a Tuesday. It's a computer, you bought it, and it does what you tell it — in your room, with the door shut.

Buying hardware from a company you'd never heard of this morning is a leap. These are the terms that make it a small one.
Plug it in, connect your keys, push it hard for a month. If it isn't what you wanted, it goes back and we refund it. One email, no form, no restocking fee.
Your key, your provider, your invoice. We take no cut, add no markup and run no proxy — the box talks to Anthropic or OpenAI exactly the way your laptop already does.
The hardware is a one-time purchase and it works fully without paying us anything again. There's no tier that unlocks the thing you just bought.
There is no analytics to switch off, because there is none to begin with. The box has no reason to report on you and it doesn't.
One payment. One box. Your keys, your provider, your invoice — and a machine that keeps working after you've gone to bed.
In stock · 30 days to send it back
No, and it doesn't have to be. The box runs 7–8B models locally, which handle the everyday middle of the work — summarising, extraction, classification, drafting, first-pass code. For anything harder you point it at Claude, GPT or Gemini on your own key and get exactly the frontier model you're already paying for. The local side isn't the ceiling, it's the option for documents you'd never upload.
Still unsure? Ask us directly — a person answers.
Get the setup guide — which providers plug in where, what an 8 GB local model actually handles, and a monthly note on what people are automating with theirs.
One email a month · unsubscribe in one click