stratoai.pro // ai-directory // hermes-agent
HM
// AI agents · assistants and browser

Hermes Agent

Open-source (MIT) agent from Nous Research with its own memory and skills, on your own server.

Vendor · Nous Research ↗
Pricing
Free (MIT) · optional Nous Portal (Free $0, pay as you go; plans from $20/mo) · model and hosting extra
Model
Open source
Evaluation
External evaluation
Verified
November 16, 2026
◆ Operator verdict

Hermes Agent is an open-source (MIT) AI agent from Nous Research that you install on your own computer or server: it remembers what it learns, writes its own skills and answers you from Telegram, WhatsApp, Slack or the terminal. The software itself costs nothing: you pay for the AI model, the machine it runs on and the time it takes to lock it down properly, and I can set it up for you and train your team. What the homepage won’t tell you: it’s still on version 0.x, on a Mac it’s only supported on Apple Silicon, and how safe it is depends on how you set up approvals.

+ What it’s GOOD for

  • Businesses that want their own assistant in Telegram, WhatsApp or Slack without paying a software license
  • Teams with recurring processes that want the agent to turn them into reusable skills
  • Anyone who needs to choose and switch AI models, or run a local one, without getting locked into a vendor
  • Scheduled tasks in plain language, like the daily report, the weekly review or the nightly backup
  • OpenClaw, Claude Code, Codex or Cursor users who want to bring over their settings, memories and skills with the official importer

− What it’s NOT for

  • Teams with nobody to maintain a server and keep it updated: it’s self-hosted software and still on version 0.x
  • Intel Macs: they aren’t supported
  • Budgets that need a fixed price: model costs depend on usage, and Nous Portal lists its prices in dollars
  • Leaving it unattended with approvals turned off (off mode): you remove the safeguard against dangerous commands
  • Anyone who doesn’t want to install or maintain anything: Hermes Cloud exists inside Nous Portal, but a managed agent like Grok Bot or Manus is a better fit

◇ Where it fits across StratoAI lines

AI strategist at night next to a mini PC acting as a server, with a terminal open on the laptop and a monitor showing a diagram of one agent connected to several messaging apps

Hermes Agent is an open-source AI agent you install on your own computer or server: it remembers what it learns, writes its own skills (reusable procedures) from the work it does and answers you from Telegram, WhatsApp, Slack, Discord, Signal, email or the terminal. It’s built by Nous Research, released under the MIT license, and the software is free: what you pay for is the AI model you pick and the machine it runs on. I’ve checked it against its official sources. Here’s what it is, what it really costs and where I wouldn’t recommend it.

What Hermes Agent is (and what it isn’t)

Nous Research describes it in its repo as “the self-improving AI agent built by Nous Research”: an agent that gets better the more you use it. Three things define it:

  • It’s yours. You install it on your machine, on a VPS or in a container. It isn’t a website you subscribe to. Nous offers optional hosting (Hermes Cloud) inside the paid Nous Portal plans, but you don’t need it to use the agent.
  • It learns. It keeps a memory, creates skills after complex tasks and improves them as it uses them. More on that below.
  • It lives where you already talk. A single process, the “gateway,” connects it to your messaging apps, so you can text it from your phone while it works on a server.
If you’re looking for…What it isWhere to go
Hermes Agent (Nous Research)A self-hosted, open-source AI agent with its own memory and skillsThis page
HermèsThe French fashion houseNothing to do with AI
hermes.soA domain that’s for saleNot the product (more on that under fake sites)
An agent that works in the cloud with nothing to installA different kind of productGrok Bot or Manus

Who makes it: Nous Research

It’s built by Nous Research, Inc., and you can check that in three official places: the GitHub repo, which carries a “Built by Nous Research” badge; the footer of the official site, which reads “© 2026, Nous Research, Inc.” next to the MIT license; and the official domains page in its docs.

Project facts, checked against the GitHub API on November 16, 2026:

FactValue
LicenseMIT
Repo createdJuly 22, 2025
LanguagePython
Stable releasev0.23.1, published on GitHub on November 10, 2026; a patch to v0.23.0 “The Hearth Release” from November 5
Release pace“Canary” test builds almost every day
Stars and forks263,637 stars and 59,437 forks

It’s still on the 0.x series: a very active, very widely used project that hasn’t shipped a 1.0 yet. I factor that into everything I recommend below.

How it works: its own memory and skills

What sets Hermes Agent apart from a chatbot with tools is what Nous calls a “closed learning loop”:

  • Memory the agent curates itself. Two text files in ~/.hermes/memories/: MEMORY.md, with up to 2,200 characters of what it has learned, and USER.md, with up to 1,375 characters about you. Both are loaded into the prompt at the start of every session. I like that they’re short: it forces the agent to keep only what matters.
  • Skills it writes itself. After a complex task it can write a skill and refine it as it uses it. Skills follow the open agentskills.io standard.
  • Search across past conversations, using full-text search (FTS5) with model-written summaries, so it can recall things from one session to the next.
  • A model of you as a user, built with Honcho, that sharpens session after session.
Open notebook on a wooden desk with a hand-drawn loop linking a brain, a note, a toolbox and a magnifying glass, next to a fountain pen, round glasses and a monitor showing two open text files
The learning loop: memory, skills, search across past sessions, and around again.

You can use it from a full terminal interface and also from the Hermes Desktop app, available for macOS 12 or later and for Windows (through the Microsoft Store or an .exe installer).

What you can do with Hermes Agent

  • Talk to it from your apps. Telegram, Discord, Slack, WhatsApp, Signal and email, plus the terminal, all from one gateway. The official messaging gateway guide adds many more: SMS, Home Assistant, Mattermost, Matrix, Microsoft Teams, LINE, iMessage (through BlueBubbles), DingTalk, Feishu/Lark, WeCom, Weixin, QQ, Yuanbao, ntfy and the browser. Nous’s comparison page sums it up as “more than 20 messaging platforms.”
  • Schedule tasks in plain language. It ships with a built-in cron: the daily report, the nightly backup or the weekly audit, delivered to whichever platform you choose.
  • Split work across subagents that run in parallel.
  • Run it wherever you want. Six built-in execution backends: local, Docker, SSH, Modal, Daytona and Vercel Sandbox, plus Singularity through a plugin. It isn’t tied to your laptop: you message it on Telegram while it works on a cloud machine.
  • Connect MCP servers to give it more tools. If that’s new to you, here’s what MCP servers are.
  • Use any model you like. Nous Portal (300+ models according to the README and docs; the plans table on the site says 200+), OpenRouter, OpenAI (including ChatGPT sign-in for Codex), Anthropic (with an API key or with Claude Max plus extra credits), z.ai/GLM, Kimi, Fireworks, Nebius, your own endpoints and local models through Ollama or vLLM. You switch with hermes model, no code changes.

◆ My advice for getting started. Don’t hook up ten channels and twenty tools on day one. One channel (Telegram or Slack), one model with a spending cap and one or two scheduled tasks you can measure. Once that works, let the agent build up skills on real work.

What Hermes Agent really costs

The official answer is in the site’s FAQ, under “How does pricing work?”: “Hermes Agent is free and open source under the MIT license. Model providers and optional hosted services have their own pricing.”

Here’s how the bill breaks down:

ItemDo you pay?How much
The softwareNoFree, under the MIT license
The AI modelYes, unless you stick to free models or run a local one on your own hardwareDepends on the provider and your usage. There’s no single figure, and I haven’t verified per-model pricing
Where it runsIt dependsOn your own machine, nothing beyond the hardware. On a VPS, whatever your provider charges: the README mentions “a $5 VPS” as an example, but I haven’t checked any hosting prices
Nous PortalOptionalFree plan at $0; paid plans from $20/mo (table below)
Setting it up, securing it and maintaining itYes, in timeYour time, or the time of whoever sets it up for you. It’s the cost people forget most

Nous Portal plans

Nous Portal is Nous Research’s optional subscription. It bundles model access and, on paid plans, a “Tool Gateway” for web search, image generation (FAL), text-to-speech and a cloud browser (Browser Use), billed against your subscription, into one account, so you don’t have to collect five separate API keys. You turn it on with hermes setup --portal. Prices from the official site, checked on November 16, 2026, in dollars:

PlanPriceWhat’s included
Free$0, pay as you goFree models only; no hosted tools or cloud agent
Plus$20/moShared usage, free and paid models, hosted tools and 1 Hermes Cloud agent (4 GB RAM, 25 GB disk)
Super$100/mo5x Plus usage; cloud agent with 4 GB RAM and 100 GB disk
Ultra$200/mo10x Plus usage; cloud agent with 8 GB RAM and 100 GB disk

◇ Price in euros, VAT and per-token costs: not confirmed. The site points to Nous Portal for current plans and billing, and I couldn’t confirm how taxes are handled or a final price in euros. Per-token cost doesn’t have a fixed figure either: it depends on the model and the provider you choose.

How I’d budget for it: as three separate lines. First, the model, with a spending cap at the provider if it offers one. Second, the machine: your own computer for testing; for a business, a machine or container dedicated to the agent. Third, the hours for setup and maintenance. You’ll find estimates on Google of hundreds or thousands of dollars a month; I couldn’t verify them, and they depend entirely on the model and usage, so I wouldn’t take them at face value.

How to install Hermes Agent on Windows, Mac and Linux

Official support according to the platform support page, checked on November 16, 2026:

SystemSupport
macOSApple Silicon only. Intel Macs aren’t supported: there’s no Intel desktop package; the same page says you can install the CLI there, but outside of support
WindowsNative Windows 10 and 11 (x86_64 and ARM64). The MSIX package requires Windows 11 22H2 or later
Linux and WSL2x86_64 and aarch64
DockerSupported, with the official nousresearch/hermes-agent image on Docker Hub
Nix and Android (Termux, aarch64)Tier 2 support

The Linux desktop package is disabled in published releases, so on Linux you use it from the terminal.

Linux, macOS and WSL2:

curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Windows (PowerShell):

iex (irm https://hermes-agent.nousresearch.com/install.ps1)

Then:

  1. hermes setup runs the full setup wizard. If it finds OpenClaw, Claude Code, Codex or Cursor on your machine, it offers to import their settings, memories and skills, with a preview first.
  2. hermes model to pick a provider and model.
  3. hermes gateway setup and hermes gateway start to connect Telegram, Slack, WhatsApp and the rest.
  4. hermes doctor if something breaks and hermes update to update.

Copy the commands from the official site, never from a blog or a video: that’s exactly the gap fake sites exploit.

Security: what it ships with and what’s up to you

The security docs describe eight layers:

  1. User authorization and DM pairing.
  2. Approval for dangerous commands, with three modes: smart (the default), manual and off.
  3. A denylist of paths it can’t write to.
  4. Container isolation.
  5. Credential filtering for MCP servers.
  6. File scanning for prompt injection.
  7. Isolation between sessions.
  8. Input sanitization.

The sensitive one is number two: with approvals.mode: off, the safeguard against dangerous commands is gone and the agent runs them without asking.

Hand holding a phone showing an approval request from the agent with two buttons, red and green, next to a mini PC server, a laptop with a terminal open and a notebook with a padlock sketch
Sensitive actions go through approval: the agent asks and you decide from your phone.

My minimum rules:

  • Never off on an agent that touches customer data or production systems.
  • A dedicated machine or container, not someone’s laptop.
  • Only authorized users can talk to it; everyone else is locked out.
  • Least-privilege keys and a spending cap at the model provider.

Watch out for fake sites and accounts

The docs themselves warn about impersonation: fake “Hermes” websites, GitHub accounts that mass-mention users and emails offering “credits.” These are the official domains:

  • hermes-agent.nousresearch.com (site, docs and installers)
  • portal.nousresearch.com (Nous Portal)
  • nousresearch.com
  • The NousResearch organization on GitHub and the nousresearch/hermes-agent image on Docker Hub
  • hermes-assets.nousresearch.com (desktop packages, updates and installer mirrors)

That same page warns that there’s no Hermes Agent or Nous Research cryptocurrency token: if someone offers you one, it’s a scam.

hermes.so isn’t the product: it’s a domain that’s for sale (“Domain Name for Sale”), which I checked on November 16, 2026. Today’s search results also turn up landing pages, copies of the docs and hosting services that aren’t on that list. They may be legitimate, but they aren’t Nous Research, so don’t hand over your keys without knowing who’s behind them.

◇ Correction. An earlier version of this directory entry credited Hermes Agent to a company that doesn’t appear in any official source, linked to that for-sale domain and described it as a closed browser-agent SaaS. That was wrong. I’ve corrected it using the official sources listed at the end.

Hermes Agent vs OpenClaw

This is the comparison people search for most, and the two have more in common than you’d think: both are open source under the MIT license, and both run on your own infrastructure.

Hermes AgentOpenClaw
Who makes itNous Research, Inc.The OpenClaw Foundation, an independent foundation (openclaw.ai)
LicenseMITMIT
GitHub stars (Nov 16, 2026)263,637394,534
Software priceFreeFree
What you pay forModel and hosting; optional Nous PortalModel and server; its site says there’s no subscription and no hosted tier (“No subscription. No hosted tier.”)
What sets it apartCurated memory, skills it writes itself, cron and six execution backendsSee the OpenClaw review
Switching between themImports your OpenClaw persona (SOUL.md), memories, skills, command allowlist, messaging settings and allowlisted API keys (with --migrate-secrets), through the official agent-importers plugin and hermes import-agent openclaw—

How I choose: if your team already runs OpenClaw and it works, don’t switch just because something new is out. If you’re starting from scratch and want an agent that builds up its own procedures over time, Hermes Agent is my first pick. And if you’re on OpenClaw and curious, the official importer’s --dry-run shows what it would bring over before anything moves. Nous publishes its own comparison, but take it for what it is: it’s written by the maker of one of the two.

Hermes Agent alternatives

I checked Grok Bot, Claude and Manus pricing on their official pages on November 16, 2026 (Manus from its help center), and it’s in dollars; confirm before you buy. The OpenClaw details were checked on its site the same day.

AlternativeWhat it isEntry priceNote
OpenClawOpen-source (MIT) assistant you run on your own serverFree; you pay for the model and the serverNo hosted tier or subscription, according to its site. OpenClaw review
Grok Bot (SpaceX Super Intelligence and Cursor)Proprietary agents with a cloud computer that work in your apps$30/mo (Premium)You can’t choose the model. Grok Bot review
Claude Cowork (Anthropic)Proprietary agent inside the Claude app, with a desktop appClaude Pro, $20/mo billed monthlyClaude review
ManusGeneral-purpose cloud agentPro from $20/moTaken from the official help center; the pricing page couldn’t be read. Manus review
Claude Code (Anthropic)Coding agent in the terminalPaid Claude plan or APIBuilt for code, not for staffing a messaging channel. Claude Code review

If you want something that works with nothing to install, look at Grok Bot or Manus. If you want the agent, its data and the model under your control, the real comparison is Hermes Agent versus OpenClaw.

My operator’s verdict

Hermes Agent is one of the self-hosted agents that makes the most sense today for a business that wants its own assistant: no license fee, you choose the model, you talk to it from the apps you already use, and over time it builds up memory and procedures that reflect how you work. That’s why it’s the first one I’d suggest you try.

What I won’t hide from you: it’s still on version 0.x and ships builds almost daily, so it needs someone to maintain it; on a Mac it’s only supported on Apple Silicon; and security isn’t solved out of the box: that part’s on you, through approvals, authorized users and where it runs. My recommendation: start with one channel, one model with a spending cap and one scheduled task. If it isn’t saving you real work within two weeks, it’s not the tool for you.

I’ll set it up and teach you how to use it

I’ll set up Hermes Agent for you and train the team who’ll use it. The work comes down to choosing where it runs, connecting only the channels and tools that are needed, setting approvals properly, picking the model with its spending cap and leaving your team using it without depending on me.

Hermes Agent FAQ

What is Hermes Agent?

An open-source AI agent from Nous Research that you install on your own computer or server. It has its own memory, writes its own skills and can be reached from messaging apps and the terminal. It has nothing to do with the fashion brand Hermès.

What does the Hermes agent do?

It carries out tasks for you using whichever AI model you choose: it works in the terminal with its tools, schedules recurring jobs, splits work across subagents and answers you on Telegram, Discord, Slack, WhatsApp, Signal or email. It also saves what it learns and writes skills so it does repeat work better next time.

Do you pay for Hermes Agent?

Not for the software. The official site puts it this way: “Hermes Agent is free and open source under the MIT license.” What you pay for separately is the model provider and any optional hosted services, such as Nous Portal.

How much does it really cost to run Hermes Agent?

The software is free. You pay for the AI model (depending on the provider and your usage, or nothing if you stick to free models or a local one on your own hardware), the machine it runs on and, if you want it, Nous Portal: Free at $0 with pay-as-you-go usage, Plus at $20/mo, Super at $100/mo and Ultra at $200/mo, in dollars. On top of that, budget the time to install, secure and maintain it.

What can I do with Hermes Agent?

Scheduled daily reports, weekly reviews and nightly backups set up in plain language; a team assistant in Slack or Telegram; long tasks that run on a server while you get on with your day; and processes the agent turns into reusable skills. With MCP servers you can plug in more tools.

How do I set up a Hermes agent?

On Linux, macOS (Apple Silicon only) and WSL2, run curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash. On Windows 10 or 11, open PowerShell and run iex (irm https://hermes-agent.nousresearch.com/install.ps1). Then hermes setup walks you through everything. There’s also a desktop app, Hermes Desktop, for macOS and Windows, and an official Docker image.

Does it work on an Intel Mac?

It isn’t supported. The official docs say macOS on Intel processors isn’t a supported platform, and there’s no Intel desktop package; according to the platform support page you can install the CLI, but outside of support. On an Intel Mac, the sensible option is to run it on another machine, like a Linux server or a container, and talk to it from your apps.

Do I need an account to log in to Hermes Agent?

No. Hermes Agent has no account or website to sign in to: you install it and open it with the hermes command or with Hermes Desktop. The exception is the hosted agent (Hermes Cloud), which you manage from Nous Portal. You only sign in to the services you connect, like Nous Portal (with hermes setup --portal) or your ChatGPT or Claude account to use their models. If a website asks for your details to “log in to Hermes,” check that it’s on the official domains list.

Who makes Hermes Agent?

Nous Research, Inc. That’s what the GitHub repo, the official site’s footer and the official domains page in its docs all say. There’s no other company behind it, and hermes.so is a domain for sale that has nothing to do with the product.

Which AI models can I use with Hermes Agent?

Pretty much any: Nous Portal (hundreds of models), OpenRouter, OpenAI (including ChatGPT sign-in for Codex), Anthropic (API key or Claude Max with extra credits), z.ai/GLM, Kimi, Fireworks, Nebius, your own endpoints and local models through Ollama or vLLM. You switch with hermes model.

Hermes Agent vs OpenClaw: which should I choose?

Both are open source under the MIT license and run on your own infrastructure. If you already use OpenClaw and it works, there’s no need to switch. If you’re starting from scratch and want an agent that builds up its own memory and skills, I start with Hermes Agent, which can also import your OpenClaw settings, memories and skills. Neither one charges for the software.

Is it safe to let it work on its own?

That depends on how you configure it. It ships with eight security layers, including approval for dangerous commands, container isolation and user authorization. But if you set approvals to off, it runs commands without asking. My rule: approvals on, a dedicated machine or container, and authorized users only.

Does it work with WhatsApp and Telegram?

Yes. From a single gateway it connects to Telegram, Discord, Slack, WhatsApp, Signal and email, plus the terminal, and the messaging guide lists more than twenty platforms in all. You configure it with hermes gateway setup and start it with hermes gateway start.

What is Nous Portal, and do I need it?

It’s Nous Research’s optional subscription: access to hundreds of models and, on paid plans, a Tool Gateway with web search, images, voice and a cloud browser, with a $0 Free plan (pay as you go) and paid plans from $20 to $200/mo, in dollars. You don’t need it: you can use your own API keys or a local model.

What are virtual AI agents?

Programs that use an AI model to carry out tasks for you, not just answer questions: they run commands, use tools, follow up on work and report back. Hermes Agent is one you host yourself; others, like Grok Bot or Manus, run in the vendor’s cloud.

What are the top 5 AI agents?

There’s no universal top 5: it depends on whether you want to host it yourself or have it managed for you. The five I compare on this page are Hermes Agent, OpenClaw, Grok Bot, Claude Cowork and Manus, with Claude Code if what you need is a coding agent.

Can you help me roll it out at my company?

Yes. I’ll set it up for you and train the team that’s going to use it. You can see how I approach agent implementation or message me directly.


Official sources checked on November 16, 2026: Hermes Agent repo and README · GitHub releases · Official site, plans and pricing FAQ · Official domains · Platform support · Installation · Model providers · Memory · Security · Tool Gateway · Migrating from OpenClaw · Nous comparison page · Nous Portal · OpenClaw repo. Alternative pricing (November 15, 2026): Grok Bot · Claude · Manus (official help center).

[ Why StratoAI ]

StratoAI is a cross-industry AI implementation firm. An active portfolio of clients in real estate, specialist retail franchises, restaurant groups (AI work, not hospitality consulting), architecture, B2B distribution, private clinics and D2C ecommerce. A real production stack: Claude Code, n8n and agents running 24/7 — no demos, no slide decks.

A real case, anonymized under NDA: Port concession — diagnosis, business plan, brand identity, renders and website delivered in 15 days and submitted five days before the concession deadline. Executed with Focus Sprint™, the flagship methodology.

Want to bring Hermes Agent into your operations?

If this tool fits your case, I’ll implement it inside one of StratoAI’s four lines: Signals, Automations, AI Agents or MicroSaaS. And if it doesn’t fit, I’ll tell you straight. A 15-minute Contact Call, no strings attached.

// Frontier stack

Models I evaluate · providers I run in production

Frontier + open-weight stack · continuous benchmarking · no vendor bias

Want to talk? I reply personally John Guerrero · StratoAI founder