Hermes Agent By Nous Research: Who Makes It, What It Does, And The Models To Run Inside It

Hermes Agent by Nous Research is a free, open-source AI agent that remembers what you tell it, writes its own skills, uses real tools and runs jobs on a schedule.

Nous Research is the AI lab that builds it, and the code is on GitHub under the MIT licence.

Below you get who the company is, how the Hermes models differ from the Hermes Agent, what the agent does, how to install it and what it costs.

After that, I use live GoldieBench scores to show how the models you can plug into it actually perform.

I checked every fact on 12 October 2026, and I name the source for each one.

Hermes Agent by Nous Research at a glance

Who Nous Research is

Nous Research describes itself on its website as "a pioneer in open AI training and research".

Its stated mission is to "create and proliferate open access to intelligence".

The company is best known for publishing open-weight models that anyone can download from Hugging Face.

Its GitHub organisation was created in May 2023, and its first Hermes model, Nous-Hermes-13b, was uploaded in June 2023.

The Block reported in April 2025 that the company was founded in 2023.

The same report named the founding team as CEO Jeffrey Quesnelle, Karan Malhotra, a pseudonymous researcher called Teknium and Shivani Mitra.

The Block and Fortune both reported a $50 million Series A round led by Paradigm in April 2025.

The Block also reported about $20 million of seed funding before that round.

I have left out anything about people or funding that I could not find in those reports.

The Hermes models versus the Hermes Agent

Nous Research uses the name Hermes for two separate products.

A model is the brain that writes the answer.

An agent is the system around the brain, which adds memory, tools, a schedule and a way to reach you.

This matters on a benchmark site, because a model can be scored on its own and an agent cannot.

Hermes Agent works with models from Nous Portal, OpenRouter, OpenAI or any endpoint you point it at.

What Hermes Agent does

The README calls it "the self-improving AI agent built by Nous Research".

Memory lets the agent keep notes between sessions and search its own past conversations.

Skills are procedures it writes for itself after a hard task, and the docs say they improve during use.

The docs list more than 60 built-in tools, covering web search, browsing, vision, image generation and speech.

One gateway process connects the agent to more than 20 messaging platforms, including Telegram, Discord, Slack, WhatsApp, Signal and email.

Scheduled jobs run plain-English tasks on a timer and deliver the result to any of those platforms.

Subagents let it split a large job across isolated helpers that work in parallel.

Profiles let one machine run several separate agents, each with its own model, memory and keys.

Plugins add extra powers from a curated catalogue you can search with hermes plugins search.

The README lists seven terminal backends, which are local, Docker, SSH, Singularity, Modal, Daytona and Vercel Sandbox.

Hermes Desktop is the official app, and Hermes Cloud is the official hosted agent.

My sibling post on the Hermes Agent backend covers the backends in more depth.

How to install Hermes Agent

The easy route is Hermes Desktop, which you download from the Hermes website.

The terminal route on Linux, macOS or WSL2 is curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash.

On native Windows, the official PowerShell command is iex (irm https://hermes-agent.nousresearch.com/install.ps1).

Run hermes setup to choose a model provider, or hermes setup --portal to log in to Nous Portal in one step.

Run hermes doctor to check the install.

Run hermes model whenever you want to change the brain.

Run hermes profile create to make a second agent with a different model.

I keep one profile per model, which is how I compare brains without touching my main setup.

My best Hermes Agent setup post ranks six full builds.

What Hermes Agent costs

The Hermes Cloud page said you need at least $2 in credits, or an active subscription, to deploy a Cloud Agent.

You can run everything for nothing with a free Portal model or a local model.

Free models rotate and have rate limits, so I use them for testing.

How fast Hermes Agent ships

GitHub listed 37 tagged releases between v0.2.0 on 12 March 2026 and v0.21.6 on 8 October 2026.

The v0.2.0 notes say the project went from a small internal project to a full agent platform in just over two weeks.

The Herald Release notes list about 3,650 commits from more than 650 contributors.

That release added streaming voice you can interrupt, signed webhooks, agent-to-agent messaging and cited research.

My own install printed a v0.21.4 build when I ran hermes --version for this post.

Is Hermes Agent safe?

Hermes Agent is as safe as the access you give it.

The official security page describes eight layers of defence, including command approval, deny-by-default chat access, container isolation and scanning of project files for prompt injection.

A hard blocklist stops catastrophic commands, and the docs say no setting overrides it.

The same page warns that file-write guards are defence in depth and not a hard boundary.

I keep approvals on, and I start every new agent on read-only work.

What is official and what is independent

Hermes Agent, Hermes Desktop, Hermes Cloud and Nous Portal are official Nous Research products.

Herald OS is an independent open-source desktop that uses Hermes Agent as its interface. FlyHermes is an independent paid host run by HF Data Studio LLC.

Both say on their own pages that they are not affiliated with Nous Research.

The names overlap because version 0.20 of Hermes Agent was called the Herald Release.

Hermes Agent compared with other agents I cover

My best Hermes Agent alternative post covers each of those.

Who Hermes Agent is for

Hermes Agent suits founders, operators and agencies who want one agent that learns their business and works on a schedule.

Hermes Desktop makes it reachable for beginners.

It is a weaker fit if you only want coding help inside an editor.

It is also a weaker fit if you want software that never changes, because the version number is still below 1.0.

The models behind Hermes Agent

Hermes Agent does not ship its own model, so the agent itself has no score on this site.

GoldieBench is my leaderboard, where every model gets the same one-shot build prompts and each build is rendered and scored from 0 to 10.

That tests the brain on building things, which is only one of the jobs an agent does.

It does not test memory, scheduling or tool chains, so treat these numbers as a guide to build quality only.

I should be plain about what is missing.

Nous Research's own Hermes 3, Hermes 4 and Hermes 4.3 models are not on the board, so I have no score for them.

The MissingNo stealth model on Nous Portal is not on the board either, and my Hermes MissingNo post explains what I saw when I tried it.

Paid brains you can run in Hermes Agent, scored

Hermes MoA is the one entry that comes from Hermes itself.

It is the Mixture of Agents feature, where a panel of models answers and an aggregator writes the final reply.

You configure it with hermes moa, and it sits close to the top of that table.

MiniMax M3 is the value pick, because it scores within a third of a point of the leader at a much lower input price.

Free brains for Hermes Agent, scored

LongCat-2.0 has the higher number, but only 4 tasks are scored, which is why the board marks it provisional.

Laguna XS 2.1 has a full set of tasks and a low average, so I would not give it a hard build.

Free models are fine for summaries, research briefs and simple tool calls, which is most of what a daily agent does.

The local models page lists the open-weight models I have run on my own Mac.

How I would choose a brain for Hermes Agent

Start on a free Portal model so you learn the agent before you pay for anything.

Add a strong paid model as a second profile when you need real builds.

Keep a fallback with hermes fallback add, so a rate limit does not stop your work.

My best Hermes Agent model post and my best LLMs for Hermes Agent post go deeper on the choice.

Also On Our Network

• Agent Operatorsthe step-by-step Hermes Agent install and setup walkthrough
• AI Income Deskwhat a business actually gets from Hermes Agent
• AI Tool Verdictevery Hermes Agent feature reviewed and scored
• agentos.guidethe seven-step Hermes Agent quick-start for your Agent OS
• agentos.guideJulian's guide to the Hermes v0.20 Herald Release

Learn the agent on a free model first, then use the board to choose the brain you run inside Hermes Agent by Nous Research.