Hermes Agent Just Got 10X Better...

[00:00:00]Hermes' agent just got 10 times better.

[00:00:02]How long do you actually sit there waiting for your AI to say the first word? 4 seconds? 5? Long enough to pick your phone up? That dead air just got deleted. Hermes' agent shipped its biggest update yet, and barely anyone has covered what it unlocks. Because plugged into my agent OS, this stops being a chat window and starts being a team. Hey, I'm the digital avatar of Julian Goldie, and I help people learn AI tools and actually use them in their work instead of just watching demos.

[00:00:24]Today, I'm walking you through what changed in Hermes' agent, why it matters more than it sounds, and how I run Hermes' agent inside my agent OS. Stay to the end because the last part is the piece almost nobody sets up, and it's the reason my system keeps getting better every time I use it. So, here's what happened. On the 20th of July, News Research shipped Hermes' agent version 0.19, the Quicksilver release. Since the last version, there's over 1,000 merged pull requests, around 3,300 issues closed, and more than 450 community contributors on a free open-source agent. And the headline is speed. Cold start used to eat about 4.3 seconds before your first message even reached the model. Now, it's about 0.9 seconds.

[00:01:02]That's roughly an 80% cut, and it applies everywhere: the command line, the gateway, the terminal interface, the desktop app, and scheduled runs.

[00:01:09]Here's why I care about a few seconds.

[00:01:11]Inside my agent OS, I run Hermes' agent as the writer on my SEO station, and I use that to draft a full content series answering the questions AI Profit Boardroom members ask me most. I'm not firing off one of those, I'm firing off dozens. When every single run started with a 4-second stall, that stall quietly ate my afternoon. Now, the content I'm building for the AI Profit Boardroom starts moving the second I hit enter, and that changes how much I'm willing to attempt in a day.

[00:01:35]But speed off the line is only half of it. Reasoning models now stream their thinking live by default. So, instead of staring at a spinner for 30 seconds wondering if the thing is broken, you watch it work. The response also paints token by token instead of line by line, so it feels alive the whole way through.

[00:01:49]That sounds cosmetic. It isn't. When you see the reasoning as it happens, you catch a bad plan in 5 seconds instead of waiting on a finished answer that went the wrong way. The desktop app got a proper performance pass, too. Streaming markdown is around 14 times faster, big file changes render smoothly instead of choking, and switching sessions is snappy again. If you run long conversations, that's the difference between a tool you enjoy and a tool you avoid. Then there's my favorite quiet upgrade. Small approvals are on by default now. When a command gets flagged, a model reviews it first instead of throwing every action back at you manually. You get the safety net without babysitting the screen. Hermes agent can also pull secrets straight from Bitwarden or 1Password, so your keys live where keys should live instead of scattered across config files.

[00:02:27]Background agents got sturdier, too. You can watch sub-agents work with live transcripts, so you're not guessing what's happening. And there's a new delivery ledger, which sounds boring and is brilliant. It records the finished responses it's being sent, and if the gateway crashes before delivery lands, it retries on the next boot, so work your agent already did doesn't vanish.

[00:02:44]There's also profile routing. One gateway can send different channels or threads to completely different profiles, each with its own settings, skills, memory, and secrets. Now, if you want the exact system I'm running all of this inside, that's the Agent OS, and I've packaged the whole thing up in the AI Profit Boardroom. You get the full Agent OS zip file ready to install, so you're not rebuilding my dashboard from scratch off a screen recording. We've built a complete 30-day roadmap to go with it, so you know what to set up first, what to plug in second, and what to leave alone until you're ready. There are walkthroughs covering how I wire Hermes agent into the Agent OS, how I set up profiles for different models, and how the memory layer is structured.

[00:03:20]And if you get stuck, we run live calls four times a week where you can share your screen and ask about your actual setup instead of guessing. Every question that comes in, I answer with a video tutorial. It's aiprofitboardroom.com, and the link is in the comments and description. Okay, so let's talk about why the Agent OS matters more than any single update. Here's the trap almost everyone falls into. You've got the smartest assistant in history, and its entire job is waiting for you to type.

[00:03:43]It can't start anything on its own. It can't check on your competitors overnight. It can't draft the outreach while you're on a call. The moment you stop typing, it stops working. You are the bottleneck. You're the courier carrying answers from one tab to another. The Agent OS fixes the organization problem, not the model problem. Same Hermes agent underneath, arranged completely differently. In my Agent OS, Hermes agent is the commander.

[00:04:04]Underneath it, I've got over 30 profiles in around 14 stations, and each one has a job. There's a radar station that pulls the latest AI and automation news every 24 hours, so I'm never guessing what's worth covering. There's a voice profile I can talk to out loud in real time, and it talks back, and it builds while we're talking. There's a competitor monitor that runs around the clock. There's an outreach station that handles email campaigns from one command. There's an SEO station where I drop in a keyword, generate the piece, and publish it to the site in a single click without logging into anything. And there's a studio for images, video, and voice, plus a game studio that builds a playable game from a single prompt.

[00:04:38]Here's a real one. I use the Agent OS to map out the entire onboarding flow new AI Profit Boardroom members walk through on day one. Welcome messages, what they see first, which walk through to open, what the next step is after that. One workflow planned end-to-end. New AI Profit Boardroom members land inside and know exactly what to do instead of scrolling around trying to find the starting line. Now, here's where the army stops being a metaphor. Missions fan out, a squad leader plans the job, then pushes tasks out to different agent profiles through a Kanban board. On my content board, there's a director who decides how a piece should be made, a builder who makes it, and a judge who quality controls the output before it moves. So, I'm not sitting in a chat waiting for one answer, I'm dropping the task in, and the work lands on the board finished. And this is exactly where the new update earns its keep inside my Agent OS. Fanning out 10 missions only works if you trust what's happening while you're not looking, and that's what live sub-agent transcripts fix. I can open a background job mid-run and read what it's actually doing, so a job that's drifted gets caught early instead of landing on the board as finished rubbish. The delivery ledger covers the other end of it. If the gateway falls over after the work is done, that finished response is recorded and goes out on the next boot instead of disappearing. And because smart approvals review flagged commands on their own now, a background mission doesn't see it frozen at 2:00 in the morning waiting for me to tap approve.

[00:05:49]Faster starts, visible work, nothing lost. That's what makes running missions in parallel actually practical rather than just impressive. And I know the obvious objection here, 10 agents running at once just makes 10 times the mess, right? That's exactly why the memory layer matters. Every exchange saves into an Obsidian vault that all my agents read from, my brand, my voice, my projects, my standards. So, the 10 background jobs come back sounding like one team that knows the business, not 10 strangers who've never met. That memory layer is also the thing I'd set up first if I was starting again. Because right now, if you finish a session in one tool and then open a different one, that second tool has no idea what you just did. You re-explain yourself every single time. With the vault, the context is already there and the whole system gets sharper every time I use it instead of resetting to zero. I use that same setup to build a full content pipeline that brings the right people into the AI Profit Boardroom. Topics, hooks, scripts, captions, follow-ups, all planned in one pass, all built from the same memory of what the AI Profit Boardroom actually is and who it's for.

[00:06:45]That's the part you can't fake with one clever prompt. One more thing worth noting, you don't need a wall of new tools to run this. I plug in models I already have access to, plus free open models, and they all become profiles I can switch between. New model drops, I plug it in, it joins the lineup. Nothing gets wasted, nothing sits in its own tab, and everything answers to the same command line.

[00:07:03]So, that's it. Hermes agent just got dramatically faster off the line. It streams its reasoning live, it recovers work it used to lose, and it handles approvals and secrets like a grown-up system. And the Agent OS is what turns all of that into a team that runs missions in parallel, remembers everything, and keeps improving instead of starting from scratch every morning.

[00:07:21]If you want the full process SOPs and 100 plus AI use cases like this one, join the AI Success Lab. Links in the comments and description. You'll get all the video notes from there, plus access to our community of 85,000 members who are crushing it with AI. And if you're about to go and actually build this, here's what I'll tell you. The install isn't the hard part. The hard part is the second day, when you're deciding which profiles to create, which models to point to which station, how to structure the memory vault so your agents don't contradict each other, and how to set up approvals so background missions don't stall waiting on you.

[00:07:51]That's exactly what the AI Profit Boardroom is built for. You get my full Agent OS zip file ready to install, the 30-day roadmap so you know the order to do it in, the prompts I use for the profiles, tutorials on wiring Hermes agent into the stations, and live coaching calls four times a week where you can share your screen and get unstuck in minutes instead of days.

[00:08:10]There are over 4,000 members in there building the same way. Come and join us at aiprofitboardroom.com.