Is Hermes Agent free? What it costs to run, self-hosted or hosted

Yes, the Hermes Agent software is free: it's an open-source AI agent from Nous Research under the MIT license, per its GitHub repository. Running it costs money for the AI model, a machine that stays on and your upkeep time. Nous Portal's paid plans run $20 to $200 a month, and pay-per-token API use has no fixed ceiling.

Most of the pages that rank for this question are setup guides that stop at "the software is free". That's true and it leaves out the bill. This piece covers each cost the way you'd meet it, with today's prices from the vendors' own pages, then lays out who should self-host and when paying someone to host it is worth it.

What's free and what isn't

The software is free to download, run, change and redistribute. The Hermes Agent GitHub repository lists the MIT license, and the Hermes Agent documentation covers installs for macOS, Linux, Windows and Android, plus a desktop app.

Three things around it aren't free:

• The model. Hermes doesn't ship with a model. The providers page says you need at least one way to connect to an LLM, and lists dozens: Nous Portal, OpenRouter, Anthropic, OpenAI and local servers like Ollama.
• A machine that stays on. An agent that runs jobs while you sleep needs a computer that's awake. That's your own Mac left running, a rented server or a hosted cloud agent.
• Your time. Someone installs it, connects the chat apps, wires in email and calendar, updates it and fixes it when something breaks.

The model is usually the biggest and least predictable line, so it comes first.

What the AI model costs

The model bill depends on which route you pick. There are four main ones, and the docs are clear about how each one bills.

Pay per token with an API key. You add a key from Anthropic, OpenAI, OpenRouter or another provider, and each call is billed by the tokens it reads and writes. Anthropic's model pricing page lists Claude Sonnet 5 at $2 per million input tokens and $10 per million output tokens, Claude Opus 5 at $5 and $25, and Claude Haiku 4.5 at $1 and $5. An agent uses more tokens than a chat window would for the same request. The Hermes providers page warns that it "may make several model calls per user turn". Set a spending cap with whichever provider you use before you leave it running overnight.

Free models. Free routes exist, and they come with limits. OpenRouter's limits page caps free model variants at 20 requests a minute and 50 requests a day until you've bought at least 10 credits, then 1,000 a day. Nous Portal has a $0 tier with free models only and standard rate limits. For an agent that runs scheduled jobs, OpenRouter's 50 requests a day goes fast.

A local model. The Hermes docs include a guide to running Hermes with Ollama and open-weight models "entirely on your own machine", with no API cost. The trade is hardware: a local model needs a machine with enough memory to run it well, and it's worth testing whether the model you pick handles long chains of tool calls before you rely on it.

A subscription. Nous Portal sells a Hermes Agent subscription: Plus at $20 a month, Super at $100 and Ultra at $200. Each includes Hermes Agent usage within limits that reset daily, weekly and monthly, and each includes hosting for one cloud agent (up to 4 GB of RAM and 25 GB of disk on Plus). The Nous Portal docs say one sign-in covers 300+ models plus web search, image generation, text-to-speech and browser tools.

Two subscriptions people already pay for come up a lot. Per the providers page, Claude Max can sign in to Hermes, but Hermes draws only from extra usage you've bought on top of Max, never from the base Max allowance. Claude Pro can't use that sign-in at all; Pro users need an API key billed per token. Anthropic's pricing page lists Max from $100 a month. A ChatGPT plan can be connected through Nous Portal for supported inference without a paid Nous subscription, and Nous's pricing page notes tools and cloud hosting are charged separately in that case.

What it costs to keep it running 24/7

An agent that answers you on Telegram at midnight or sends a morning briefing needs a machine that's on. You have three ways to get one.

Your own computer. Free if you already own it, as long as it stays awake and online, including through OS updates while you're away. A spare desktop on a shelf, kept plugged in and awake, is the simplest version.

A rented server. The Hermes docs say it can live on "a $5 VPS". Today's list prices are close to that. DigitalOcean's Droplet pricing starts at $4 a month for 512 MiB of memory, $6 for 1 GiB and $12 for 2 GiB. Hetzner Cloud lists shared cloud servers from €5.99 a month, excluding VAT. Most of the cost is the setup on the server: Linux, a process that restarts after a crash, firewall rules, and signing in to OAuth providers from a machine with no browser, which the Hermes docs handle with SSH port forwarding.

A hosted cloud agent. Nous Portal's paid plans include one cloud agent, and its homepage says server costs for bigger sizes or extra agents bill to your Portal balance. Serverless backends like Daytona and Modal are another option the docs name, described as costing "nearly nothing when idle".

The cost most guides skip: your time

The software moves quickly. The GitHub repository showed new commits landing the same day we checked it, which is good for features and means updates are a regular job.

Here's what the time goes into, roughly in the order you'll meet it:

• Installing it, choosing a model provider and getting keys or subscriptions sorted.
• Connecting a chat app so you can reach it from your phone.
• Wiring in email and calendar, so it knows who you met and what's waiting on you.
• Writing or installing skills and scheduled jobs that do repeat work.
• Updating it, reading what changed, and fixing whatever an update or an expired login broke.
• Backing up its memory, so a dead server doesn't take months of context with it.

If you enjoy that work, it's free. If you'd rather spend the hours on your company, price them. Our audit uses $150 an hour for a founder's time, so if upkeep takes you 2 hours a month, that's 2 x $150 x 12 months = $3,600 a year. Put your own hours and rate into the same line.

Self-hosted or hosted: the full comparison

Here's every route side by side, with prices from each vendor's page today. Your own time isn't in the table; add it from the section above.

Route	Plan price	Model	Machine	Who keeps it running	Source
Your Mac + a local model	$0	$0 per token; needs enough memory	Your own Mac, left on	You	Hermes docs
Your Mac or a VPS + free models	$0	$0, capped at 50 or 1,000 requests a day on OpenRouter	$0 to about $12 a month	You	OpenRouter limits, DigitalOcean
A VPS + an API key	$0	Per token, e.g. Claude Sonnet 5 at $2 in and $10 out per million	About $4 to $12 a month	You	Anthropic pricing, DigitalOcean
Nous Portal Plus	$20 a month	Included within daily, weekly and monthly limits	One cloud agent included	You set it up; Nous hosts the agent	Nous Portal
Hockeystick Personal	$49 a month	$10 of AI a month on Claude Opus, or your own key	Included	We host and keep it running	Hockeystick pricing page
Hockeystick Team	$199 a month, unlimited people	$100 of AI a month, or your own key	Included	We host and keep it running	Hockeystick pricing page

None of these is the right answer for everyone, and the cheapest row on paper isn't always cheapest once you count hours.

Who should self-host

Self-host if you like the terminal, want full control over where every file lives, or want to run local models for privacy. You'll learn how the agent works, and you can change anything. A small VPS plus an API key with a spending cap is a sensible start, and Nous Portal is the shortest path if you'd rather not juggle several accounts.

Self-hosting also makes sense when the agent is for you alone and the work is mostly coding, research or personal tasks, where a broken feed costs you an evening and nothing else.

When a hosted plan is worth paying for

Hosted is worth it when the agent has to know your company and run unattended, and nobody on your team wants to be its system administrator.

Hockeystick is a company brain with a hosted AI agent that works from it. You chat with your agent in the free Hermes app, or in Slack or Telegram once we connect it on your setup call. Nous Research doesn't make or endorse Hockeystick; we host the agent and add the parts around it. Per our pricing page, Personal is $49 a month and includes:

• The brain: a note for every person, meeting, decision and doc, with Ask for answers that cite their sources, and the Network of everyone you work with.
• Your own agent on Claude Opus, hosted and running around the clock, with 41 skills from the morning briefing to contract review.
• Hands-off filing: once Gmail and Google Calendar are connected, your agent files new people, meetings and decisions into the brain and sends you a receipt.
• Approvals: it asks before it sends, posts, deletes or spends.
• $10 of included AI a month, or your own Claude, OpenAI or OpenRouter key, which it uses first.
• A free 30-minute setup call, daily backups with a copy off the server, and one live office-hours call a month.

Per the same pricing page, Team is $199 a month flat for unlimited people. Everyone reads the brain, asks it questions and talks in its threads, with private spaces and roles. If you'd like the idea behind the brain first, we wrote up what a company brain is.

Hosted costs more than DigitalOcean's $6 Droplet. It's priced against the hours above: the setup, the email and calendar wiring, the backups and the fixes. If those hours are worth more to you than Personal's $49 a month, hosted wins. If they're a hobby, self-host and enjoy it.