Is Hermes Agent free? Nous Portal, Hermes Cloud and bring-your-own-key costs

Hermes Agent is free, and the part that costs money is the model behind it. The FAQ on its own site says Hermes Agent is "free and open source under the MIT license", and the homepage offers both the desktop app and the terminal install with no price attached to either. Your bill therefore comes from one of three decisions: bring your own provider API key and pay that provider, subscribe to Nous Portal and spend its credits, or run a local model through Ollama and pay nothing per token at all. If you want the short recommendation, a self-hosted Hermes Agent pointed at Ollama on hardware you already own really is zero a month, and the cheapest paid route with hosted tools included is Nous Portal Plus at $20.

What follows is what each of those costs as of 23 October 2026, taken from Nous's own pages rather than from any summary of them. One figure is missing on purpose, and the section on Hermes Cloud says exactly why.

#What is free and what is not?

The licence covers the software, not the inference. That division is the whole answer to the question in the title, so it is worth setting out before the prices.

The homepage puts the same point in one line: "Model providers and optional hosted services have their own pricing." Nothing on that list is a licence fee, which is why "is Hermes Agent free" has a straight yes for an answer and a longer answer about running costs.

#What do Nous Portal's tiers cost?

Nous Portal is the first-party option: one subscription that covers model inference and the hosted tools, billed as credits rather than per token. There are four tiers, and the credit figures are higher than the prices because each paid tier adds a 10 percent bonus.

Three things that table does not show, all from the same page. Credits can be topped up outside the subscription in fixed amounts of $10, $20, $50, $100 or $200 through Stripe. Credits are spent on three different things, not just models: model usage across the catalogue, the hosted tools priced per use, and the server cost of Hermes Cloud hosting. And there is a route that sidesteps the subscription, because the Portal says you can connect an eligible ChatGPT account for supported inference without a paid Nous subscription, with tools and cloud hosting billed separately.

The rollover cap is the figure the Portal gives for how much unused credit carries forward, and the page states it as a cap without explaining the mechanism further. The useful consequence is the ratio: on Plus you can bank at most $10 of the $22, so a month you barely use does not fund a month you lean on.

On how many models a subscription reaches, Nous's pages do not agree, and this post will not pick one for you. The Portal's tier cards say "200+ Models", while the models section further down the same page is headed "300+ models". The Hermes Agent documentation for the Portal advertises "300+ frontier models, one bill", listing named families from Anthropic, OpenAI, Google, DeepSeek, Qwen, Kimi, GLM, MiniMax, xAI and NVIDIA alongside Nous's own Hermes models, then "280+ additional models". Those are two different claims on the vendor's own site, so check the catalogue inside the Portal if the exact count decides your purchase.

#What does Hermes Cloud cost?

This is the one price in this article that cannot be quoted, and the reason is worth stating plainly rather than hedging. Hermes Cloud is the always-on hosted option, for running an agent without keeping a machine awake at home. Its own page on the Portal does not list rates: it tells you to "View current instance sizes and pricing in Nous Portal", so the per-instance figures sit behind a sign-in. Searching Nous's sites for the instance pricing itself, rather than for the page it ought to be on, turns up the same pointer.

What that page does publish is the billing model, in enough detail to work out the shape of a bill before you commit to one:

• An instance "is billed from your Nous credit once each day (UTC), in arrears, based on what it actually used that day".
• While running, compute is "metered to the second of actual uptime", plus storage.
• While stopped, you pay "storage only, so your data and setup persist between sessions".
• There is a "$2 minimum in credits, or an active subscription, to deploy a Cloud Agent".
• "Inference and tool usage are billed separately, on top of the instance price."

So a cloud agent is a per-day instance charge plus storage plus whatever inference and tools it consumes, all drawn from the one credit balance a subscription tops up. The practical planning note is that stopping an instance does not stop the bill, it reduces it to storage, and that the $2 floor means credits, not a card, are what gate a deployment.

#How do you run Hermes Agent for nothing?

Point it at a model on your own machine. Hermes documents this route itself as a zero API cost setup using Ollama, with no API keys and no per-token charges. The install and configuration are four commands:

curl -fsSL https://ollama.com/install.sh | sh
ollama pull gemma4:31b
hermes setup
hermes

ollama --version confirms the install, and hermes gateway runs it as an always-on bot rather than an interactive session. The catch is not money, it is memory, and the model table is where the real decision sits.

Only the largest of those supports tool calling, which is what an agent needs to do anything beyond talk, so the free route in practice means gemma4:31b and the hardware to hold it. The guide gives 8 GB of RAM as the floor for a 3B model and 32 GB or more as the recommendation for 27B and up, with 5 GB of free storage as a minimum and 30 GB or more if you keep several models. No GPU is required, an NVIDIA card with 8 GB or more of VRAM "speeds things up significantly", and a CPU-only setup works but answers more slowly.

#Which option should you pick?

Setting up the Portal route takes one command on a fresh install, hermes setup --portal, which handles the sign-in, model selection and the Tool Gateway; on an existing install hermes model lets you add Nous Portal as a provider while leaving the ones you already have in place. Either way the credentials land in ~/.hermes/auth.json, separately from config.yaml.

For how Hermes Agent compares with the other self-hosted option people weigh it against, see Hermes Agent vs OpenClaw. For the same local-model trick on OpenClaw, which has the same hardware arithmetic, see running OpenClaw free with Ollama. And for what Hermes Agent is, what it runs on and how its skills work, start at our Hermes Agent hub.

Prices and tiers were read from Nous's own pages on 23 October 2026. Subscription pricing and the Hermes Cloud instance table are both things vendors change without a changelog, so check the Portal before you budget against the figures here.