Hermes AI: The Fastest Growing Agent Ever

Hermes is now the fastest growing AI agent of all time and I need you to understand what that means. So Cryis Beta posted the numbers today. OpenClaw, the tool that took the AI world by storm, went from zero to 40,000 GitHub stars in 61 days. Hermes just did it in 45 and in just the last seven days alone Hermes gained three times more stars than OpenClaw.

That has never happened before with any AI agent. So what is Hermes and why is everyone switching? More importantly, what can you actually do with it? So let me break all of this down for you.

So number one, Hermes agent is built by Noose Research. They're a real AI lab based in Austin, Texas founded in 2023 by Jeffrey Quesnel, Karen Maholtra and Technium and Shivani Mitra. They raised a 50 million dollar Series A led by Paradigm. They're the team behind the Hermes model family which has been downloaded over 50 million times.

They have about 20 employees and they've been building open source AI tools since before most people knew what an AI agent actually was. So this isn't some random hobby project that got lucky on social media, this is a well-funded research lab that built something people actually want. Now here's why people are moving to Hermes so fast. The number one complaint about OpenClaw, and I hear this constantly, is that it struggles remembering, it struggles improving and quite often context is a big issue with any of these agents.

So you start a new session, you work for a problem, you close the window and the next day it struggles to remember the context, the conversation, the improvements, the feedback that you give it. So quite often it really struggles that even though OpenClaw does have a file based memory system, it's got sole MD, it's got memory MD, but you have to set up manually, you have to maintain it yourself and quite often it's just not as efficient as you feel it could be. There's a lot of back and forth, there's a lot of repeating yourself and it can be quite messy. Now Hermes solves that problem at the core level.

It remembers everything, every session, every task, every project and it does it automatically. They've actually added a new memory system in recently where you can plug in and plug out and cross out across all of your different agents. For example they added a native Honcho memory option but you can also use Obsidian as well, that's a good option too. The agent itself handles it on its own using something called a closed gap learning loop.

Here's how the learning loop works in plain English. So you give Hermes a task and it does a task then, and this is a part that actually matters, it looks at what it just did, these steps, and basically turns that into a reusable skill, saves that skill and solves the whole thing in memory. So the next time you give it a similar task it loads a skill it already made and does it faster so there's no extra prompting from you, no setup, it just gets better. Jeffrey Quesnel, the CEO of Noose Research, showed this off by having Hermes write a 79,000 word novel across multiple sessions.

No human editing was done between those sessions. The agent remembered the characters, the plot, the style, everything. It just kept going and this learning loop changes the economics of using an AI agent entirely because every hour you spend with Hermes makes it more useful, right? With most AI tools hour 1 and hour 100 can often feel the same or like you're going backwards sometimes even, right?

With Hermes hour 100 is dramatically better than hour 1. Now I want to talk about what you can actually do with this thing because this is where it gets very practical. Hermes has over 40 built-in tools, web search, browser automation, code running, image generation, file handling, text-to-speech. They added one recently where it can actually create videos with diagrams in a kind of explainer animation and it also has scheduling so you can set up scheduled tasks and cron jobs and everything else and it connects over 200 AI models through Noose portal, open router, open AI, or you can set up your own local setup with something like Olama and you can pick which brain powers your agent and you can switch anytime with one command so there's no locking as well.

But the feature most people miss is the gateway and I think this is actually the real product. So you connect Hermes to Telegram, to Discord, to Slack, to WhatsApp, Signal or email and one single gateway processes and handles all of them together and then the agent just lives there. Now the cool thing about this is it's basically online even if you close your laptop so you can pick up the same conversation on your laptop later and it remembers everything across platforms. Now of course OpenClaw, you can message it from WhatsApp, you can message it from Telegram and everywhere else but the problem is it's not in the same way, it's not learning, it's not detailed, it's not documented and of course you could set up OpenClaw to do this but it's not at the core level.

If you look at the architecture of Hermes versus OpenClaw, OpenClaw is really a good agent, it's powerful but at the same time it's not going to be learning at its core architecture. You're gonna have to build that in and then that's gonna have to be a skill that you're reminded to use and then it's gonna forget that skill and then this context just drops and drops and drops and drops and the performance drops as well. So with Hermes it's like having a team member that never locks off essentially. So for example you could tell Hermes to check five competitor websites every morning and just send you a summary in Telegram at 8am.

That's literally what I did and you said that once in plain English and it runs every single day without you touching it ever again or you could have it monitor your emails, pull out the important stuff, send you a digest in WhatsApp every afternoon. One setup runs forever. Now if you want to learn how to set up Hermes workflows like this to save time and get more leads for your business, that's exactly what we cover inside the AR Profit Boarding. We've already built 30-day roadmaps around tools like Hermes step-by-step with tutorials showing you exactly how to set up automated workflows, connect it to your messaging apps and run background tasks that bring in leads whilst you sleep.

You get four coaching calls every week where you go deep on stuff like Hermes. Link in the comments description or go to the ARProfitBoarding.com to get access. Now let's talk about why the growth numbers are so wild here. Hermes hit the number one spot on GitHub trending.

Technium, the head of post training at Noose Research, posted about it. So they got 47,200 stars, 347 contributors, over 6,000 forks and 1,400 merged pull requests and they're shipping a new major version almost every single week. They're now on version 0.8 in two months and here's something that tells you everything about where the momentum is heading. Hermes has a built-in migration tool for open core users so you literally type one command and Hermes clog and migrate is a command that you write inside terminal and then it'll pull over your settings, your memories, your skills, your API keys, pretty much everything.

And that's a company that's so confident in their products they built a direct pipeline from their biggest competitor. One user on Substack actually wrote that they installed Hermes on the same server they're running open core on, right, and it took about 11 minutes. Within two hours the agent had created three skill documents from tasks they gave it. It completed a similar research task 40% faster using those skills.

So there was no prompt tuning, there was no messing around going back and forth with feedback, right. The agent just taught itself and then learned it from itself, right. And that's a key difference. With open core, when you want the agent to get better, you, the human, have to write better skills.

You, the human, have to write better prompts. You, the human, have to write better instructions. And quite often it breaks. We've all seen that with open core, right.

With Hermes, the agent writes his own skills. He improves them during use and it stores them permanently. Now let me talk about who is actually behind this because the team matters too. So Noose Research, as I mentioned before, isn't just building an agent, they're building the models themselves.

The Hermes model family, the Nomos models, the Psyche models, they built Atropos, a reinforcement learning framework for training AR models on real tools and real tool use data, right. And here's a clever bit. Every time Hermes agent completes a task in the real world, that trajectory data, meaning the exact sequence of tool calls and decisions the agent made, can feed back into training the next version of the model, right. So the agent creates data and the data trains better models and then the better models make a better agent.

And it's a flywheel that creates a positive feedback loop. And that's something no other open source agent has right now. You know, you've got open core that has a great system, but at the same time they're not building models to go inside it, right. Now Peter Steinberger, the creator and the founder of open core, has moved over to OpenAI, right.

But at the same time he's not in control of those models, right. And let's be honest, he's probably not got much insight into what's going on over there because they're a big company and they've got all sorts of crazy stuff going on. With Hermes, they're building models directly alongside their actual agent. So it's totally different.

Now they also built something called distro, distributed training over the internet. The idea is you can train AR models across regular consumer computers instead of needing massive data centers. And that's a massive research bet on decentralization. And it ties directly into the whole philosophy.

AI that belongs to you, not a corporation, right. It's open source. Now let me get into some of the specific features that make Hermes different in day-to-day use. First of all, profiles.

As of version 0.6, you can run multiple completely separate instances of Hermes on the same machine. And each one has its own memory, its own skills, its own personality. So you could have one Hermes instance that handles your marketing research, another one that manages your customer follow-ups, and another one that does your content repurposing, right. They all run at the same time.

They don't cross-contaminate. And second, scheduled automations, right. You tell Hermes what you want done and when in plain English. And it has a built-in cron schedule, very similar to OpenCLR in that way.

And it delivers the results to whatever platform you want. So you can put it into Telegram or Slack or email or wherever. No code, no cron syntax, you just say what you want. And third is sub-agents.

So Hermes can spawn isolated mini-agents for parallel tasks. So if you give it three things to research, it doesn't do them one at a time. It spins up three sub-agents, runs them all at the same time, and brings you back the combined results. And that's a huge saver on complex jobs.

And fourth, sandbox options. So you can run Hermes locally on your laptop, in Docker, over SSH to a remote server, or on a cloud GPU cluster, or through serverless platforms like Modal or Daytona. And the serverless option is smart because your agent just hibernates when you're not using it and wakes up when you need it. So you pay almost nothing when it's idle.

And fifth, security. And this is where Hermes has a real edge. You know, OpenCLR has had some rough patches, right? There was a critical vulnerability, CV2026, that allowed token theft through a single malicious link.

There were prompt injection issues, and SNYK found over 1,400 malicious skills in OpenCLR's skill marketplace. Hermes ships with safer defaults. So you've got command approval flows, dangerous pattern blocking, memory scanning for prompt injection, read-only root file systems, namespace isolation, and zero publicly reported agent-level security breaches so far. Now, I don't want to make the sound like OpenCLR's bad, because it's not.

It's a great tool. Really powerful. But you can kind of see why Hermes is getting the edge and why it's growing so quickly. And also bear in mind, OpenCLR also has NVIDIA's NemoCLR Enterprise wrapper.

It has a massive ecosystem. And for some use cases, especially if you need that platform reach or that volume of prebuilt skills, OpenCLR is still a great pick. But the direction is clear. The trend line is moving toward agents that learn.

Agents that improve themselves. Agents that don't make you do all the maintenance work. And the speed of Hermes development is something that I've personally never seen before in open source. You know, let me just run you through the release timeline really quick here.

So version 0.2 shipped in late February. Version 0.3 in mid-March. Version 0.4 two weeks later. 0.5, the security hardening release, came at the end of March.

0.6, added profiles and MCP support. 0.7, added pluggable memory providers and the Camoflox anti-detection browser. And now 0.8, just dropped two days ago. And that added background task notifications, live model switching across all platforms, and free MIMOv2 Pro on-noose portal.

There's eight major releases in about six weeks. We've 209 merged pull requests in the latest release alone. And 82 resolved issues. And that pace is unusual even for the fastest moving AI companies.

Now, here's why I think this matters most for anyone watching this, right? The age of set-it-and-forget-it AI agents is coming, right? You tell an agent once what you need, it does it, it remembers how, it gets better at it, and it runs in the background while you go and live your life. And that's the promise of Hermes.

And they're shipping fast enough that the promise is actually turning into reality. Version by version, week by week. The people who start learning this stuff now, the ones who figure out how to set up these automated workflows, how to connect agents to their business operations, how to build systems that run without them, are going to have a massive advantage over the next 12 months, right? And the people who wait, they're just going to be scrambling to catch up.

So here's what I'd like you to do, you know, if I was you. So go look at Hermes, the GitHub is newsresearch forward slash Hermes hyphen agent. You can just Google it, you can install it, try connecting it to Telegram or WhatsApp, give it a simple recurring task, something like monitoring a topic or summarizing emails, watch it learn, watch it build skills, and you'll see why 47,000 people have already started this thing in less than two months. And if you want step by step help setting this up inside your business, with coaching calls, with people who are already running Hermes to automate lead generation content and save time, come join us in the AI profit boardroom, right?

We've got a dedicated Hermes setup roadmap inside there right now. We've got daily tutorials, walking you through exactly how to connect it to your business apps, how to use a memory system, how to build automations that bring in clients on autopilot. We've got 2800 members inside there. A lot of them are already using Hermes and sharing their workflows and a prompt library built around agent automation and a member map so you can find and connect with other Hermes users need you and get help anytime of the day.

Link in the comments description or go to the ARprofitboardroom.com. This is moving fast, it's not slowing down and Hermes is just getting better and better.