Hermes Agent v0.5 Just Changed AI Agents Forever

Hermes agent v0.5 just changed agents forever. Hermes agent v0.5 dropped yesterday, March 28th and this is the biggest update yet. I'm going to break down exactly what's new, why it matters and what you should do right now if you're running a business and want an AI agent that actually works for you around the clock because this release is a turning point. Here's the thing most people don't get about Hermes agent.

Every other AI tool you've tried resets. You open it, you explain your business, you explain your clients, your tone, your preferences and then the session ends and it forgets all of it. Tomorrow you start over again. Hermes doesn't work like that.

It lives on a server, it remembers everything, it builds skills from experience. The longer you run it the better it gets at exactly how you work and v0.5 just made it significantly more capable, more reliable and this is the part people are going to care about. Way more connected to the models you actually use. Let me show you what's inside.

The headline upgrade in v0.5 is model access. Massive. New portal, Hermes own inference portal now supports over 400 models through a single endpoint. One connection over 400 models you switch with one command and then there's the new Hugging Face integration.

Hugging Face just became a first class provider inside Hermes for integration with the Hugging Face inference API. A curated model picker built specifically for agentic tasks and a live endpoint probe that checks which models are actually available right now. Why does that matter? Hugging Face hosts thousands of open source models, many of them free or near free to run.

Being able to plug Hermes directly into that and have it automatically pick the right model for the job, that's a genuine unlock for anyone running an AI automated business and watching the API costs. For example, like an agency owner could run entire workflows through Hermes on free or cheap Hugging Face models and keep costs close to zero. Now before I keep going, if you want coaching on exactly how to set up Hermes agent for your specific business, generate more leads of it and get it running automated workflows for you, the link to the AI Profit Boardroom is in the comments in the description. We've got a four-weekly coaching course specifically covering AI automation setups like this, daily tutorials, 30-day roadmaps to get results fast and a 2,600 business owners already doing this.

Go to the airprofitboardroom.com and get in. The next big thing in 0.5 is what they're calling telegram private chat topics. This sounds small, it's actually huge. You can now run multiple isolated projects inside a single telegram chat.

Each topic, each project thread has its own skills bound to it, its own context, its own memory. Think about what that means in practice. You've got one Hermes agent, you're messaging it on telegram, Monday morning you're in the client A topic handling their content pipeline. Then you switch to the sales outreach topic and it knows you're in a completely different workflow, different tools, different context, different memory.

One agent, multiple businesses or projects, all isolated or persistent. For anyone juggling more than one client or more than one revenue stream, which is basically every solo business owner I know, this is exactly what was missing. That's another upgrade in this release that I want to flag as well about trust. So Noose Research did a full supply chain audit in v0.5.

They removed a compromised dependency, a package called lightllm, pinned all dependency versions, regenerated the locked file of hashes and added a CI workflow that automatically scans every single pull request for supply chain attacks. Now why would I bring this up? Because it sounds technical right? Well this is a tool that can run on your server with access to your files, your messaging apps, your API keys, your client data.

The security of that matters and the Hermes team showed that they take this seriously, that's worth noting. Here's another one that's actually relevant to anyone running Hermes with Claude or OpenAI models. Claude and Hermes have fixed the anthropic output limits. Previously it was hard coded at 16,000 tokens max, meaning if you were doing long tasks with Claude, responses would get cut off, truncated.

Now it uses the actual per model native output limits. Claude Opus 4.6 gets 128k tokens, Claude Sonic 4.6 gets 64k. That's 8x more output per task than before. If you've been running Hermes with Claude and wondering why long tasks were getting short, that's fixed.

And the GPT model is worth mentioning too. GPT models were sometimes describing what they intended to do instead of actually doing it, taking action. Hermes v0.5 added specific guidance that forces GPT models to actually use tools rather than just talk about them. That sounds like a small bug fix, but if you've watched an AI agent narrate its way through a task, instead of completing it, you know exactly how frustrating that is.

Now it's fixed. Let me zoom out for a second because I want to understand the trajectory here. I want to talk about that. So Hermes agent launched in February 2028, six weeks ago.

Since then it's gone from a small internal project to 15,200 github stars, nearly 2,000 forks, and a community shipping hundreds of pull requests per release cycle. This is not a slow-moving project. v0.2 came out on March 12th, v0.3 on March 17th, and v0.4 shortly after, and now v0.5 on March 28th. Five major versions in six weeks, each one adding significant functionality.

The pace is the signal. This is the fastest moving open source AI agent right now, and it's free to use. Compare that to OpenClaw, which is a strong project, but moves on a corporate timeline. It costs money to run, and this is a key difference.

It doesn't learn from your specific workflows. Every session with OpenClaw, you're working with a tool that gets buggy, breaks, forgets a lot. It doesn't really build knowledge about how you work. It doesn't create skills based on experience.

You have to manually add them in, and Hermes actually does that, and with every release, that advantage widens. Now let me also quickly cover the plugins lifecycle hooks update, because this one matters if you want to customize how Hermes behaves. Plugins can now hook into four specific moments. Before an AI call, after an AI call, when a session starts, and when a session ends.

In plain English, you can now build or install community-built plugins that trigger specific actions at new specific points in Hermes workflows. A plugin that logs everything before every AI call. A plugin that sends you a Slack message when a session ends with a summary of what got done. A plugin that fires a webhook to your CRM when a session starts so your other tools know Hermes is active.

This is the kind of connected tissue that turns a capable AI agent into the center of a real business automation stack. And the modal SDK upgrade. So Hermes runs on serverless infrastructure, meaning it hibernates when you're not using it, and costs basically nothing whilst idle. The v0.5 update replaced an older dependency with the native modal SDK, meaning the serverless backend is now simpler, more reliable, and doesn't require tunnels to run.

You can run a persistent always-on AI agent for your entire business on serverless infrastructure for the cost of a cheap mile. And it wakes up the moment you message it. That's the economics of this tool. It runs 24-7, it remembers everything, it gets smarter over time, and the running cost is negligible.

Here's the honest bottom line on Hermes Agent v0.5. The people who are going to win with AI over the next 12 months are the ones who stop using AI as a search engine and start treating it as infrastructure. Infrastructure runs in the background. It doesn't need to be asked the same questions twice.

It knows your business, your clients, your processes, because it's running long enough to learn them. Hermes Agent is one of the most practical ways to build that right now. It's free, it's open source, it's moving faster than almost anything else in this space, and v0.5 just made it more capable, more reliable, and more connected than ever before. The gap between businesses using this kind of setup and businesses not using it is growing every single week.

The longer you wait, the more contacts and skills your competitors' agents are building up, whilst yours has nothing. If you want to get Hermes Agent running for your business the right way, not just installed, but actually generating leads, saving time, and handling real workflows, come join us in the AI Profit Boardroom. Link in the comments description, or go to theairprofitboardroom.com. We've got four weekly live coaching calls in there, walking you through AI automation setups exactly like this, daily step-by-step tutorials, 30-day roadmaps, a prompt library, and 2,700 business owners in there doing this together.

There's always someone online, and the member map lets you find and connect with people near you who are building the same sort of things. Hermes Agent v0.5. Free, open source, released yesterday, and the pace of this project tells you everything you need to know about where it's going. Don't sleep on it.

Thanks for watching.