NEW Hermes Agent V0.13 is INSANE!

Hermes Agent V0 0.3 just changed AI agents forever. So Hermes Agent just dropped version 0.13, and they're calling it the Tenacity release. That name tells you everything, because the whole point of this update is simple. The agent actually finishes what it starts.

That sounds obvious, but here's why it matters. Anyone who has used an AI agent to run tasks for their business knows the frustration. You set the agent going. You walk away.

You come back an hour later, and it's stuck, or it forgot what it was actually doing, or the whole session crashed, and now everything is just gone. You have to start over. That has been the number one problem with AI agents since day one. They're powerful, but they're also fragile.

We've seen that with all the open-claw updates, too. They quit. They lose track. They get distracted.

They break. And every time that happens, you're the one who has to pick up the pieces yourself. Version 0.13 is built specifically to fix out every major change in this release. It's about making one thing, making the agent more reliable, more tenacious, and more capable of running real tasks without you babysitting it.

Let's get into exactly what's changed it. So in one week, just one week, the Hermes team actually shipped 864 commits, merged 588 pull requests, and had 295 community contributors involved. This is a staggering pace. For context, most software teams ship updates like this over months.

Noose Research did it in seven days. Now, if you actually want to just update, all you do is you go into your terminal, like so, with Hermes already set up. And what you're going to do is just run Hermes update like that. And that's how you can update to the latest version.

So you can see, for example, I'm on version 0.13 here. Now, what does this update come with? What does it mean? Why does it matter, et cetera?

So Technium, the lead engineer and co-founder of Hermes Agent, tweeted after it dropped, finally here, what could possibly be in v0.14? Noose Research replied, can't even imagine. So when the team that built it doesn't know what's coming next, you know the momentum is real. So what is in 0.13?

The biggest thing is the headline feature, which is multi-agent Kanban. And if you're not a developer, don't let the word Kanban put you off. Here's what it actually means. In the old way, one Hermes agent did one thing at a time.

You gave it a job, it worked on it, then it moves on to the next one. It's very linear, sequential, slow. With Kanban in v0.13, you can have multiple Hermes workers run at the same time, each picking up different tasks from a shared board. So think of it like hiring a team instead of one person.

You can drop a list of jobs on a board, each worker picks one, does it, marks it done, and grabs the next one. The work moves in parallel. For a business owner, that changes how fast things can get done. If you're running a SEO workflow, for example, that involves researching keywords, writing content, formatting it, and publishing it, well, that used to be a sequential chain, one thing after another.

Now you can have separate workers handling different pieces of the same time, right? And this is the part that actually matters. So workers don't quit. The Kanban board in v0.13 has heartbeat monitoring, zombie detection, and automatic retry.

So if a worker gets stuck or it crashes, the board knows it reclaims its task and it gives it to another worker. Nothing falls through the cracks. The board also has a hallucination gate. If a worker claims it's finished something but can't actually prove it, the gate catches it before the false result gets passed forward.

This is the difference between AI that feels like impressive in a demo and AI that actually works in production. Now, if you wanna see how business owners are already using agents like Hermes Agent to automate their workflows and you want a 30-day roadmap for setting up your own multi-agent system to get more leads and run your business faster, come join us in the AI Profit Boardroom at aiprofitboarding.com. Link in the comments description. We've got coaching calls every week specifically walking you through Hermes Agent setups and a whole library of prompt templates built for exactly this sort of stuff.

Link in the comments description or go to aiprofitboarding.com. Now let's talk about the goal feature because this one might be the most underrated feature of the whole release. Here's the problem it solves. You're working with an AI agent over multiple turns, multiple back and forth messages.

You start with one job in mind but as the conversation goes on, the agent drifts. It gets pulled in different directions by new inputs. By the time your turn, your 10 turns in, it's doing something completely different from what you originally asked and that is the Ralph loop problem. Ralph being a sort of nickname for the kind of agent that keeps looping back and forth again in the original objective, spinning its wheels.

It was actually originally working with OpenClaw and that's kind of how OpenClaw developed with the Ralph loops but this is something that's even more powerful, persistent goals. So it's a well-known frustration in the agent space and forward slash goal actually fixes it. So if you want to use this inside Hermes, you just type forward slash goal followed by your target and the agent locks in, right? Across every turn, across every new input, it keeps that goal in focus.

It doesn't drift, it stays on task until the job is done. And think about what that means if you're running Hermes to run a long research task or to manage an outreach campaign or to audit your website content. You set the goal once, the agent holds it. You don't have to keep reminding it what you're doing.

You don't have to keep manually steering it back on track. It just runs. And that might sound like a small thing. It is not a small thing because every time your agent drifts off task, that's wasted time and wasted compute.

So forward slash goal removes that entirely. And now here's the next big one. And this is the one that Pasha, Daily Hermes user, he runs this agent from Telegram, called out immediately on X after the release. He actually said, the real unlock this week is a gateway restart doesn't nuke the session.

So before version 0.13, or if you updated the software, you know, if anything crashed, for example, like your Hermes gateway, or even if the connection was interrupted, your session was basically gone, right? You'd come back, you'd find nothing. And everything the agent was actually doing just got wiped, it would start over. In v0.13 session, survive restarts, right?

You see it right here. This is the update itself. And so this means the gateway goes down, comes back and automatically picks up where it left off. And for anyone running AI agents at work overnight, or running scheduled tasks through cron jobs, or just anyone who has ever lost work because the connection dropped, this is huge.

Paired with this, Checkpoints v2 rewrites how Hermes saves its state. The old checkpointing system had real problems, orphan files, building up disk space, bloating over time, state getting corrupted in weird ways. v0.13 fixes all of it, which means it has pruning, proper discard rails, no more bloat, the agent just runs cleaner and faster over time. Now let's talk about the security improvements too.

So there's a whole wave of security fixes. Eight high severity security issues got closed in v0.13. The most important one for everyday users is that redaction is now on by default. That means sensitive information, for example, like API keys, passwords, anything that looks like a credential gets automatically redacted from logs and shared debug files.

So before this, it was opt-in. Most people never turned it on, they didn't even know it existed. Now it's on for everyone automatically. They also fixed a serious Discord issue where role-based permissions could be bypassed across several different servers.

If you have a Hermes agent running in the Discord community, that one mattered. Like WhatsApp now rejects messages from strangers by default. So your agent isn't responding to random contacts, he doesn't even know. The security improvements in v0.13 matter especially if you're using Hermes in a business context.

So when you're automating real workflows, dealing with client communications, running tasks that touch sensitive data, you need to know the tool is locked down. This release moves in the right direction. Now there are also new models that were added inside here too. This was another highlight from the update.

So v0.13 adds support for DeepSeek, v4 Pro, and Grok 4.3 on OpenRouter. There's also an Al-Alpha free tier model on OpenRouter, which is worth knowing about if you want to run certain tasks at a lower cost. And Tencent's HY3 preview just got added as a paid route as well. Here's what that means practically.

You now have more options for which model handles different tasks. So you might use a cheaper model for bulk tasks and a more capable model for higher stakes work, right? Hermes gives you control over that. Each provider is now pluggable.

So you can add custom LLM providers without touching the core code, right? And that matters as new models keep releasing. There's also a new video analyze tool. And this means that Hermes can actually watch videos for you.

So Hermes can now natively understand video content on Gemini and compatible multi-model models. So you can pass a video to your agent and have it extract information, summarize it, or pull specific details for it. This is super useful for marketers, for content creators, anyone who uses video a lot. And that opens up a lot of new workflows.

Now, XAI Custom Voices is in there too. So you can now clone voices with Hermes and voice cloning support is now available as a TTS provider, text-to-speech provider, right? So if you're building voice-based agents or workflows through Hermes, for example, like customer communication, content creation, anything that involves audio, you now have access to custom voice cloning directly inside the agent, which is super powerful. There's also the Google Chat option here.

So Google Chat is now the 20th messaging platform supported. You can run your Hermes agent from Gmail, Telegram, WhatsApp, Discord, Slack, Teams, and now Google Chat. The list of places you can control your agent from keeps growing. And the idea here is that wherever you're actually working, Hermes can be there.

The dashboard also got an upgrade too. So there's a new plugins update where you can manage, enable, and disable plugins directly from the UI without touching any config files. There's a profiles management page, the analytics table is portable, and there's a new draft large theme with 18-pix base size, right? Genuinely easier to read, especially on longer sessions.

One thing that got acquired to coverage, but it's worth mentioning as well, the agent now lints its own rights. What does that mean? Well, when Hermes writes a file, a Python script, a JSON file, or YAML config, it immediately checks the output for syntax errors, right? So if something's wrong, it surfaces the error right there, rather than letting a broken file propagate forward.

For anyone using Hermes to create or edit files as part of an automation, this means fewer downstream failures from bad outputs. There's also a no agent Chrome mode now, so you can set up scheduled tasks that just run a script directly without spinning up the full agent. If the script runs clean, it's silent. If there's output, you get it.

This is useful for watchdog patterns. So checking if something is working, triggering alerts, running lightweight scheduled tasks. It's just faster and leaner than full agent Chrome jobs. SEER XNG is a privacy respecting meta search engine that lets you search the web without going through Google or Bing directly.

Having it as a native option inside Hermes gives you more control over how your agent searches the web, especially useful if you're running research workflows where you want clean, unbiased search results. Six new optional skills shipped with this release. So for example, Shopify integration. You also have, for example, all these new optional skills like so, Storefront GraphQL is now available as an optional skill.

If you run an e-commerce business, you can now connect Hermes directly to your Shopify store. There's also a shop app, personal shopping assistant skill, and a Kanban video orchestrator skill for creative workflows. The Anthropic Financial Services bundle got ported as optional final skills, and there's a SEER XNG search skill that ties into the new native SEER XNG backend, right? Seven languages are now supported for the CLI and gateway messages, as well as the Chinese, Japanese, German, Spanish, French, Ukrainian, and Turkish.

The doc site has a Chinese locale, and if you or your team just aren't native English speakers, Hermes is now meaningfully more accessible. 100 new CLI startup tips. That sounds small, but it's actually useful. Every time the CLI starts up, it shows a random tip, which is pretty interesting.

So basically the random banner at the top has 100 tips on how to use Hermes as well. So there are 100 fresh tips covering like Cron jobs, Kanban, plugins, lesser known features, et cetera. Over time, those tips add up to a pretty solid education on using the tool as well. So what does this all mean for you as a business owner or a content creator using AI automation?

Well, the honest answer is this. The gap between AI that sounds impressive and AI that actually works for your business is closing fast. A year ago, running an AI agent to handle real tasks meant constant supervision. You have to watch it, nudge it, restart it.

It was more work than just doing the task yourself. V0.13 is a meaningful step away from that. The Kanban system handles multitask parallel work. The goal command handles focus and follow through.

Session persistence handles the crash and restart problem. The security improvements handle data safety and the new models give you more flexibility for different use cases. The tenacity release isn't about flashy new capabilities. It's about reliability, right?

And reliability is a thing that determines whether an AI agent is actually useful in your business or just a tool you play with occasionally. So that's basically it. And you know, the direction is clear here. They're building toward agents that can genuinely run complex, multi-step business tasks without constant human oversight.

And if you're not using Hermes Agent or if you're using it but haven't dug into what it can actually do for your business, now is a great time to start paying attention. If you want a step-by-step tutorial on how to set up Hermes Agent for your business, how to run the multi-agent Kanban system for real lead generation and content workflows and coaching calls where you can ask live questions about your specific Hermes setup, come join us in the AI Profit Boarding. We've got a 30-day roadmap specifically around Hermes automation. Daily tutorials walking you through the exact workflows that are working right now.

2,900 business owners inside, a lot of them already running Hermes for client work, content creation and lead generation and a prompt library built around agent workflows and a local member map so you can connect with other Hermes users near you and get help 24-7. Link in the comments description or go to theairprofitboarding.com to get access. The update command, by the way, is simple. Just run Hermes update inside your terminal.

That is it. Thanks for watching. I'll see you on the next one. Cheers, bye-bye.