The cheapest way to run Hermes
Hermes itself is open source, so the real cost of running it is three separate things: where it lives, which model it calls, and how you reach it. Get each one right and a capable 24/7 agent costs remarkably little — here's the honest breakdown.
TL;DR
The cheapest Hermes stack is hardware you already own (a Raspberry Pi, an old laptop, a home server) or a free-tier host, paired with a free or local model— DeepSeek's free tier or a local model via Ollama — and your iPhone as the client. Hermes is open source, so the only unavoidable cost is model tokens, and even that can be near-zero with local or free models. Onepilot deploys that stack and drives it from your phone, with the SSH terminal, files, git, and cron included.
Build the cheapest Hermes stack and run it from your iPhone — one email when Onepilot ships.
The three costs (and how to zero out each)
1. The host — where the agent lives. Hermes needs a machine that stays on. The cheapest option is hardware you already have: a Raspberry Pi sips a few watts, and an old laptop or home server costs nothing extra. If you'd rather not own hardware, several cloud providers offer a free-tier instance big enough to run the agent — see Hermes on a VPS. Either way, the host can be effectively free.
2. The model — the only real variable. This is where money actually goes, and where you have the most control. Run a local model via Ollama on the host and inference is free (you pay only in electricity). Or point Hermes at a free or low-cost hosted model— DeepSeek's free tier landed in a recent release, and providers like OpenRouter expose inexpensive open models. Hermes reads its model credentials from environment variables, so you can switch to a cheaper model any time.
3. The client — how you reach it.Your phone. The agent runs on the host; the iPhone is just the control surface, so there's no additional hardware to buy.
Own hardware vs free-tier host
Hardware you ownhas a one-time cost and then only electricity — best if you want full control and a box that's yours. A Raspberry Pi or a repurposed laptop runs Hermes 24/7 for the cost of a nightlight.
A free-tier or low-cost VPShas nothing to buy and is reachable from anywhere, which suits people who don't want a machine at home. The trade is that free tiers can change terms, so treat the box as disposable and keep a backup of ~/.hermes/ (the skills and memory).
Both run the same agent. The cheapest absolute setup is usually hardware you already have plus a local model; the cheapest hands-off setup is a free-tier host plus a free hosted model.
Where Onepilot fits
Onepilot doesn't change the stack's economics — it makes the cheap stack easy to stand up and run. It deploys Hermes to whatever host you picked over SSH (no sudo), lets you set the model provider key and the channel in the app, and gives you the terminal, file browser, git, and cron from your phone. So you can chase the cheapest model or move hosts without re-learning a setup, and you read exactly what the agent is doing on the box you're paying (or not paying) for.
FAQ
How much does it cost to run Hermes Agent?
Hermes is open source, so there's no license cost. The real costs are three: the host (free if you use hardware you own or a free-tier cloud instance), the model (free with a local model via Ollama, or low/zero cost with a free hosted tier like DeepSeek), and the client (your phone, no extra hardware). The only unavoidable spend is model inference, and even that can be near-zero with a local or free model. A Raspberry Pi running Hermes 24/7 costs about as much electricity as a nightlight.
Is Hermes Agent free?
The Hermes agent software is free and open source. What you might pay for is the model it calls and the machine it runs on. Run a local model on hardware you own and the running cost is essentially just electricity; use a free hosted model tier and it can be zero. You only start paying meaningfully if you choose a premium hosted model or rent a larger server than the agent needs.
What's the cheapest model to use with Hermes?
The cheapest is a local model run via Ollama on the host — inference is free, you only pay in electricity, though quality depends on your hardware. For hosted models, look at free tiers (DeepSeek added one in a recent release) and inexpensive open models via OpenRouter. Hermes reads model credentials from environment variables, so you can switch models freely; Onepilot lets you change the provider key in the app.
Do I need to pay for a server?
Not necessarily. The cheapest host is hardware you already own — a Raspberry Pi, an old laptop, or a home server — which costs only electricity to run. If you'd rather not own hardware, several providers offer a free-tier instance large enough to run the agent (the model runs at the provider or locally, so the box itself can be small). Renting a paid server is optional, not required.
Does Onepilot add to the cost?
Onepilot is the app you use to deploy and drive the stack; this page is about the cost of the Hermes stack itself — the host and the model — which you control. Onepilot's role is to make the cheap stack easy: it deploys Hermes over SSH, lets you set the model and channel in the app, and gives you the terminal, files, git, and cron from your phone so you can keep the running costs low and visible.
Run Hermes for next to nothing
Drop your email and we'll send one note when Onepilot ships on the App Store.
See also: Hermes on a Raspberry Pi, Hermes on a VPS, Run Hermes on iPhone.