Hermes Agent by Nous Research is a free, MIT-licensed AI agent with memory, self-written skills, real tools and a gateway to your chat apps, and my overall score for it is 9 out of 10.
Nous Research is the AI lab that makes it, and the same company publishes the Hermes family of open-weight models.
This is the reference version, so you'll get the company facts, every feature explained and scored, the price, the limits and a comparison table.
I use Hermes every day, and I checked the facts on 12 October 2026 against the Nous Research website, the official docs, the GitHub API and the press coverage I name below.
Hermes Agent By Nous Research: The Fact Sheet
| Fact | Detail | Source |
|---|---|---|
| Product | Hermes Agent, "the self-improving AI agent built by Nous Research" | The official README |
| Maker | Nous Research, Inc. | The company website |
| Licence | MIT | The GitHub API |
| Language | Python | The GitHub API |
| Repository created | 22 July 2025 | The GitHub API |
| First tagged public release | v0.2.0 on 12 March 2026 | GitHub releases |
| Latest tagged release | v0.21.6 on 8 October 2026 | GitHub releases |
| Stars and forks | About 252,000 stars and 54,000 forks | The GitHub API on 12 October 2026 |
| Platforms | Linux, macOS, WSL2, native Windows, Nix and Android through Termux | The official docs |
| Price | $0 for the software | The licence |
Who Nous Research Is
Nous Research describes itself as "a pioneer in open AI training and research".
Its mission is to "create and proliferate open access to intelligence".
The company's GitHub organisation dates from May 2023, and its first Hermes model on Hugging Face, Nous-Hermes-13b, was uploaded in June 2023.
Here is what reputable press has reported about the business.
- The Block reported in April 2025 that Nous Research was founded in 2023.
- It named the founding team as CEO Jeffrey Quesnelle, Karan Malhotra, the pseudonymous researcher Teknium and Shivani Mitra.
- The Block and Fortune both reported a $50 million Series A round led by Paradigm that month.
- The Block reported roughly $20 million of seed funding before that round.
- The same article described Psyche, a network for training models across many separate machines.
I haven't included anything about people or money that I couldn't find in those reports.
On its own homepage, Nous also calls Hermes Agent "the most widely used open source agent harness in the world".
That is the company's own claim, and the GitHub numbers above are the only part of it I can check myself.
🔥 Want my exact Hermes Agent By Nous Research setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
Hermes Models vs Hermes Agent
The name Hermes covers two separate products.
| Hermes models | Hermes Agent | |
|---|---|---|
| Type | Open-weight language models | Agent software |
| Since | June 2023 | Public since early 2026 |
| Recent versions | Hermes 4 in August 2025 and Hermes 4.3 in November 2025 | v0.21.6 in October 2026 |
| Where | Hugging Face | GitHub |
| Needs the other? | No | No |
The models are brains you can download.
The agent is the system that gives any brain memory, tools and a schedule.
Hermes Agent works with models from Nous Portal, OpenRouter, OpenAI or any endpoint you point it at, so a Hermes model is optional.
How I Scored Hermes Agent
I score each feature out of 10 on three things.
The first is what the feature does according to the official docs.
The second is how it has behaved in my own daily use, where I have used it.
The third is how much work it takes a normal person to get value from it.
These are my own first-person scores, and they aren't a lab benchmark.
Every Hermes Agent Feature, Scored
| Feature | Score | Why |
|---|---|---|
| Memory | 9 | It keeps notes between sessions and searches its own past chats. |
| Skills and the learning loop | 9 | It writes skills after hard tasks and improves them in use. |
| Built-in tools | 9 | The docs list more than 60, covering search, browsing, vision, images and speech. |
| Gateway | 9 | One process reaches more than 20 messaging platforms. |
| Scheduled jobs | 9 | Plain-English cron with delivery to any platform. |
| Profiles | 9 | Separate agents with separate memory on one machine. |
| Plugins | 8 | A large catalogue with one-click installs, and community quality varies. |
| Hermes Desktop | 8 | A real app for people who avoid the terminal. |
| Hermes Cloud | 8 | Clear daily pricing, with model usage billed on top. |
| Voice | 8 | Streaming speech you can interrupt, added in v0.20. |
| Terminal backends | 8 | Seven options, from local to serverless. |
| Security | 7 | Strong layers, with honest warnings about their limits. |
| Stability | 7 | It ships fast, so things change under you. |
| Price | 10 | The software is free under the MIT licence. |
Memory And Skills: 9 Out Of 10
Memory is why I use an agent instead of a chat box.
Hermes keeps agent-curated notes, and the docs say it nudges itself to save what matters.
It can also search its past sessions, so "what did we decide last week?" is a fair question to ask.
Skills are saved procedures.
After a complex task, the agent can write down how it did it, and the docs say those skills improve during use.
The /learn command turns a topic or a document into a skill on request.
I've used it to turn a full book into a skill, and it's the feature that surprises people most.
Tools And The Gateway: 9 Out Of 10
The docs list more than 60 built-in tools.
Those cover web search, page extraction, browsing, vision, image generation and text to speech.
It also connects to MCP servers, which is the standard way to plug in outside tools.
The gateway is one process that connects the agent to more than 20 platforms.
The docs name Telegram, Discord, Slack, WhatsApp, Signal, Matrix, email, SMS, Microsoft Teams and Google Chat, along with several others.
I talk to my agents on Telegram, and it's the difference between a tool I open and a teammate I message.
Scheduling, Subagents And Bot Mode: 9 Out Of 10
Scheduled jobs are written in plain English and delivered to any connected platform.
I run a 6am to-do recap, a daily Japanese lesson and a morning inbox summary this way.
Subagents let the main agent hand out parts of a job to isolated helpers.
Bot Mode gives you named bots, each with its own model, memory, skills and routines, and they can work together in a group chat.
Profiles: 9 Out Of 10
A profile is a separate Hermes home with its own config, keys, memory and skills.
I keep one profile per model, so I can test a new brain without risking my main agent.
Agencies can keep one profile per client for the same reason.
You create one with hermes profile create and run it with hermes -p and the name.
Plugins: 8 Out Of 10
Hermes has a curated plugin catalogue you can search from the command line with hermes plugins search.
When I covered the directory in Hermes Desktop, it showed around 500 free community plugins with ratings, versions and a one-click install.
It loses two points because community plugins are written by strangers, and you have to read the GitHub page before you trust one.
Hermes Desktop And Hermes Cloud: 8 Out Of 10 Each
Hermes Desktop is the official app, and the Nous homepage lists downloads for Mac, Windows and Linux.
It gives you chat, settings and plugins in a normal window.
Hermes Cloud is the official hosted agent, and its pricing page was refreshingly clear.
| Hermes Cloud size | Running | Stopped | RAM | vCPUs |
|---|---|---|---|---|
| Standard | $0.51 a day | $0.03 a day | 2 GB | 2 |
| Pro | $0.99 a day | $0.03 a day | 4 GB | 4 |
The page said those prices exclude model and tool usage, and that you need at least $2 in credits or an active subscription to deploy.
It also said an agent can scale to zero when idle.
Both lose points for the same reason, which is that the model bill is separate and harder to predict.
Voice And Terminal Backends: 8 Out Of 10
Version 0.20, the Herald Release, rebuilt voice.
The agent now speaks clause by clause as it thinks, and you can interrupt it by talking.
I tested that release in my own setup, and my Hermes v0.20 guide has the details.
Terminal backends decide where the agent's commands run.
The README lists seven, which are local, Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox.
My sibling post on the Hermes Agent backend explains when to use each one.
🚀 Ready to put Hermes Agent By Nous Research to work in your business? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
Security: 7 Out Of 10
The official security page describes eight layers of defence.
| Layer | What the docs say |
|---|---|
| User authorisation | Chat access is denied by default, with allowlists and one-time pairing codes. |
| Command approval | Risky commands are checked, the default mode uses a second model to judge them, and a hard blocklist stops catastrophic commands with no override. |
| File-write safety | Protected paths are guarded, and the docs call this defence in depth. |
| Container isolation | The Docker backend drops privileges and limits processes. |
| MCP credential filtering | Outside tools only get the environment variables you allow. |
| Context-file scanning | Project files are checked for prompt-injection patterns. |
| Cross-session isolation | Sessions are kept apart from each other. |
| Input sanitisation | The docs list the cleaning of inputs as the final layer. |
I score it 7 because the design is strong and the docs are unusually honest.
They say plainly that file guards are "not a hard boundary" and that turning approvals off is for trusted environments only.
An agent that can run commands is powerful, and power is risk.
Keep approvals on, start read-only and run hermes security to audit your plugins against known vulnerabilities.
Stability And Release Pace: 7 Out Of 10
GitHub showed 37 tagged releases between v0.2.0 on 12 March 2026 and v0.21.6 on 8 October 2026.
That is a new minor version roughly every week and a half.
| Release | Date | Name |
|---|---|---|
| v0.13.0 | 7 May 2026 | The Tenacity Release |
| v0.15.0 | 28 May 2026 | The Velocity Release |
| v0.16.0 | 5 June 2026 | The Surface Release |
| v0.18.0 | 1 July 2026 | The Judgment Release |
| v0.19.0 | 20 July 2026 | The Quicksilver Release |
| v0.20.0 | 3 August 2026 | The Herald Release |
The Herald Release notes alone list about 3,650 commits from more than 650 contributors.
That pace is a strength, because bugs get fixed and features arrive quickly.
It's also why I mark stability down, because the version number is still below 1.0 and things move.
My own install printed a v0.21.4 build when I ran hermes --version for this review, so I'm a few tags behind by choice.
Price: 10 Out Of 10
| Item | Price |
|---|---|
| Hermes Agent | $0 |
| Hermes Desktop | $0 |
| Nous Portal Free | $0, free models only |
| Nous Portal Plus | $20 a month, with $22 of credits |
| Nous Portal Super | $100 a month, with $110 of credits |
| Nous Portal Ultra | $200 a month, with $220 of credits |
| A local model | $0 |
You can run the whole thing for nothing with a free model.
The Nous homepage listed a stealth model called MissingNo as free for a limited time, and I cover it in my Hermes MissingNo post.
How To Install Hermes Agent
The fastest route is Hermes Desktop from the Hermes website.
The terminal route on Linux, macOS or WSL2 is one line.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
On native Windows, the official PowerShell command is iex (irm https://hermes-agent.nousresearch.com/install.ps1).
hermes setup asks you to choose a model provider, and hermes doctor checks the result.
My best Hermes Agent setup post ranks six full builds.
🔥 Want help getting Hermes working properly? The AI Profit Boardroom has a full Hermes Agent course and four live coaching calls a week, so you can share your screen and get unstuck. You'll be learning with 3,400+ members. → Get access here
Hermes Agent vs Other Agents I Cover
| Agent | Type | Cost | Runs where | Who I'd point to it |
|---|---|---|---|---|
| Hermes Agent | Open-source personal agent | Free, MIT | Your machine, a server or Hermes Cloud | Anyone who wants one agent that learns their business |
| OpenClaw | Open-source personal agent | Free, MIT | Your machine or server | People who want a second open-source option |
| Claude Code | Paid coding and work agent | A paid Claude plan or API | Your machine | People building software and sites |
| OpenCode | Open-source coding agent | Free, MIT | Your machine | People coding on a budget |
| Hark Pro | Hosted personal agent | A free plan, then $20 or $100 a month | Its own cloud computer | People who want no setup at all |
| StarNet | Open-source agent station | Free, MIT | Your machine | Tinkerers who like local tools |
My best Hermes Agent alternative ranking reviews each of those in detail.
Official vs Independent: A Quick Check
| Product | Official Nous Research product? |
|---|---|
| Hermes Agent | Yes |
| Hermes Desktop | Yes |
| Hermes Cloud | Yes |
| Nous Portal | Yes |
| Herald OS | No, it is an independent open-source project |
| FlyHermes | No, it is an independent paid host |
Herald OS shares a name with the Herald Release of Hermes Agent, and that's a coincidence of branding, not a sign of ownership.
Hermes Agent Limits
- It is pre-1.0 software that changes often.
- The quality of the work depends on the model you connect.
- Free models have rate limits and can disappear.
- Community plugins need checking before you install them.
- An agent with command access needs a careful owner.
Who Should Use Hermes Agent
I'd recommend it to operators, agencies and founders who want one agent that learns their business.
I'd recommend Hermes Desktop to beginners and the terminal install to anyone comfortable with a command line.
I'd point someone who only wants help writing code in an editor to a dedicated coding agent instead.
My Verdict On Hermes Agent
My overall score is 9 out of 10.
It earns that for free software, a real learning loop, a huge gateway and a very fast release pace.
It drops a point for stability and for the care that security needs.
The company behind it has a public history of open models going back to 2023, which matters when you're building a business on a tool.
🚀 Want my Agent OS with Hermes already wired in? The AI Profit Boardroom includes my installable Agent OS, the Hermes trainings and four coaching calls a week, with 3,400+ members building next to you. → Join the AI Profit Boardroom
Also On Our Network
- 🌐 How to install and set up Hermes Agent, step by step
- 🌐 What a business gets from Hermes Agent, with real use cases
- 🌐 The Hermes Agent quick-start for an Agent OS
- 🌐 The models behind Hermes Agent, with live bench scores
FAQ: Hermes Agent By Nous Research
What is Hermes Agent by Nous Research?
Hermes Agent is a free, open-source AI agent built by Nous Research.
It keeps memory across sessions, writes and improves its own skills, uses more than 60 built-in tools, runs scheduled jobs and connects to more than 20 messaging platforms through one gateway.
Who is Nous Research?
Nous Research, Inc. is an AI lab that describes itself as a pioneer in open AI training and research.
It is known for the Hermes family of open-weight models on Hugging Face.
The Block reported in April 2025 that it was founded in 2023 and had raised a $50 million Series A led by Paradigm.
Is Hermes Agent free?
Yes, the software is free.
The model is the part that can cost money.
Nous Portal listed a free plan limited to free models and paid plans at $20, $100 and $200 a month, and you can also run a local model for nothing.
Is Hermes Agent open source?
Yes, the GitHub API lists NousResearch/hermes-agent under the MIT licence.
The repository showed about 252,000 stars and 54,000 forks on 12 October 2026.
Is Hermes Agent safe?
The official docs describe eight layers of protection, including command approval, deny-by-default chat access and container isolation.
They also warn that file-write guards are defence in depth and not a hard boundary.
I score security 7 out of 10 and recommend keeping approvals on.
Is Herald OS or FlyHermes made by Nous Research?
No, because Herald OS is an independent MIT-licensed desktop made by a developer called Luke The Dev, and FlyHermes is a paid host run by HF Data Studio LLC.
Both state they are not affiliated with Nous Research.
What is the latest version of Hermes Agent?
The latest tagged release on GitHub was v0.21.6, published on 8 October 2026.
The project has shipped 37 tagged releases since v0.2.0 on 12 March 2026, so check the releases page for anything newer.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I built Goldie Agency, a 7-figure SEO and link building agency based in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery", a book on link building strategies.
- I run GoldieBench, where I score AI models on real one-shot builds.
→ Get my best AI training inside the AI Profit Boardroom
Free software, a real learning loop and an honest security page are why I give 9 out of 10 to Hermes Agent by Nous Research.
