7 Hermes Agent Upgrades You NEED To Implement Today — Transcript
Full transcript
- 0:00Hermes Agents is one of the most
- 0:01powerful AI tools you can use [music]
- 0:03right now. But for most people, it just
- 0:05feels like a chatbot inside Telegram.
- 0:07I've been running Hermes every day for
- 0:09months now. Not one agent, a whole crew
- 0:11of them. And I've learned what actually
- 0:13makes a difference. So in this video,
- 0:15I'm going to show you seven tips that
- 0:16takes Hermes from a bot you chat with to
- 0:18a team that works for you. And stay to
- 0:20the end because tip number seven is my
- 0:22favorite one. Just so you know, none of
- 0:24the tools that I mention in this video
- 0:26are sponsors. These are just things I
- 0:28genuinely use. All right. So the first
- 0:30one, build a crew, not one agent. Most
- 0:33people use a single agent for
- 0:34everything. Research, writing, coding,
- 0:37all in one chat. And after a while, it
- 0:39starts mixing things up. So I split the
- 0:41work. For example, I have a main agent
- 0:43called Orchestrator who plans and runs
- 0:45the show. Then research finds things
- 0:47out, writer writes, and developer
- 0:49builds. Each one [music] has a defined
- 0:51job. And you don't need a terminal to
- 0:53set these things up. Just ask your main
- 0:54agent something like, "Create a new
- 0:56persistent agent called research. It's
- 0:58only job is to find things out and
- 1:00summarize what it finds. Give it its own
- 1:02instructions and its own memory. And
- 1:04tell me when it's ready." And that's it.
- 1:06Your main agent builds it for you.
- 1:09Repeat it for every agent you want. Or
- 1:11ask your main agent to create several of
- 1:13them at once. Now here's my favorite
- 1:15part. Make one Telegram group and turn
- 1:17on topics in the group settings. Now you
- 1:19can give every agent its own topic, like
- 1:21its own channel. So you talk to research
- 1:23directly, to writer directly, and the
- 1:25same for every agent. And you always
- 1:27know who is doing what. And because
- 1:29every topic is its own session, every
- 1:32agent keeps its own conversation. And
- 1:34you don't end up blowing up one context
- 1:36window like you would do when talking to
- 1:37all of them in one chat. The tricky
- 1:39part, out of the box, Hermes sends every
- 1:42message in that group to your main
- 1:44agents. But there's a simple workaround,
- 1:46a small plugin that sends each topic to
- 1:48the right agent. And I have a prompt for
- 1:50you for that in the description section.
- 1:52So now you've got a crew, but every
- 1:54agent in that crew runs on an AI model.
- 1:56[music] And models cost you money.
- 1:58Bringing us to tip number two, you don't
- 2:00need the most expensive model. Most
- 2:02people pick the most expensive model for
- 2:04everything, but I recently had a whole
- 2:06memory system for my AI agents built by
- 2:08MiniMax M3, which cost a lot less than
- 2:11the top models from Anthropic and
- 2:13OpenAI. Cheaper doesn't mean bad. It
- 2:15builds the whole thing. Every agent can
- 2:17run its own model. So, think about what
- 2:20each one does. A developer agent writing
- 2:22code, sure, give it a stronger model.
- 2:24But an agent that is probably just
- 2:26checking your email or summarizing an
- 2:27article probably doesn't need the most
- 2:29expensive model in the world. Put it on
- 2:31a cheaper one. And even inside one
- 2:33provider, there's a whole family of
- 2:35models. OpenAI, for instance, has big
- 2:37models like GPT-Xtra and GPT-5.6 and the
- 2:40cheaper ones like GPT-5.5 or GPT-5.4
- 2:43mini. Same company, same account. You
- 2:46don't always have to pick the top one.
- 2:48And even with the same model, there are
- 2:50different effort levels that you can
- 2:51set. You can go from light all the way
- 2:53to extra high. So, think about the
- 2:55effort that the agent needs for that
- 2:56particular task and adjust the model and
- 2:59adjust the effort level as you see fit.
- 3:01Now, here's something a lot of people
- 3:02probably don't know. If you already pay
- 3:04for a ChatGPT plan, you can sign in to
- 3:06Hermes with that account and use the
- 3:08OpenAI models without paying anything
- 3:10extra. And the same applies to Grok. If
- 3:12you have a super Grok or X subscription,
- 3:15the higher your plan of course, the more
- 3:16usage you get. And that usage typically
- 3:18works out a lot cheaper than paying as
- 3:21you go through the API. So, now your
- 3:23crew is cheaper to run. But there's one
- 3:25problem every crew runs into. [music]
- 3:27They forget things. So, tip number
- 3:29three, give your agents a real memory.
- 3:31Agents don't have long-term memory. Chat
- 3:33history fills up and when it does, the
- 3:35agent has no option than to summarize it
- 3:37to make room. This is what is called
- 3:39compaction and every time it compacts,
- 3:41details get lost. It's like a whiteboard
- 3:44that someone keeps wiping. Say tell your
- 3:46agents on Monday exactly how you like
- 3:48your reports written. Probably by Friday
- 3:51after a few long chats, that's probably
- 3:52gone. It gets summarized away. So, I
- 3:55give my crew a wiki, plain notes in
- 3:57plain markdown files on my own server.
- 4:00Every agent resets before it answers and
- 4:02writes to it after it works. And because
- 4:05they are plain markdown files, I can
- 4:06open any page, read it to learn what the
- 4:08agents are documenting, and also fix
- 4:10where I find inconsistencies myself.
- 4:12[music] But hey, you can also outsource
- 4:15this to a librarian agent like I did in
- 4:16my setup, where the agents will take
- 4:18care of this for you on an
- 4:21The best part is that it's not only your
- 4:23Hermes agent, any AI agent you use like
- 4:25Claude Code or Codex can read the same
- 4:27wiki and be up to speed in second. And
- 4:30one rule, keep your own facts separate
- 4:32from what the agents read online. Your
- 4:33decisions and your clients live in one
- 4:36place and articles they read live
- 4:38somewhere. Because someone else's
- 4:40marketing claim should never become your
- 4:42fact. This is inspired by Andrej
- 4:44Karpathy's LLM wiki concept. So, now
- 4:46your crew remembers everything, but so
- 4:48far it only works when you talk [music]
- 4:50to it. Bringing us to tip number four,
- 4:52let your agents work without you. Hermes
- 4:55can run real tasks on a schedule.
- 4:56[music]
- 4:57That's called a cron job. Not reminders,
- 4:59real work. For example, every morning at
- 5:018:00 a.m. my agent sends me what is
- 5:03called a morning brief and that contains
- 5:05a summary of my unread emails, the tasks
- 5:07I need to handle, my calendar, and the
- 5:09weather for that day. Basically, it
- 5:11preps me for the day ahead. I don't ask
- 5:13for it. It's just waiting on my phone
- 5:15when I wake [music] up. Or say you have
- 5:17an agent check your email every 30
- 5:19minutes and maybe draft a reply for you.
- 5:21Now here's a small trick. Most of these
- 5:23checks probably find nothing new. So,
- 5:25without this trick, your phone buzzes
- 5:27every half an hour just to tell you
- 5:29nothing happened. So, in every scheduled
- 5:31job, tell the agents if there is nothing
- 5:33new to report, just stay quiet. Hermes
- 5:35has a built-in way to do that, so
- 5:37nothing gets sent at all. I will also
- 5:39put the exact prompt for this in the
- 5:41description section. But for your agents
- 5:43to be able to check your emails, your
- 5:45agents actually needs a tool to give it
- 5:47access to that. Which brings us to tip
- 5:49number five, connect your tools the safe
- 5:51way. Now, to connect my agent to Gmail,
- 5:53my calendar, Notion, and other apps that
- 5:56are out there, I use Composio. It has a
- 5:58huge pool of tools you can connect to.
- 6:00And there's a free plan to start with.
- 6:02And surprisingly, you can use it for so
- 6:04much without even having to pay. You
- 6:06sign in once in your browser and connect
- 6:08the tools you want your agent to use.
- 6:10And once you connect your Hermes agent
- 6:11to it, the agent gets a key to get
- 6:14access to these tools. And you can
- 6:15cancel it anytime. The best part is that
- 6:17your password access to these [music]
- 6:19tools never touches your server. Think
- 6:21of it like giving a house sitter a spare
- 6:23key, which you can take back whenever
- 6:24you want. And they never had the
- 6:26original key. Now, three rules I follow.
- 6:28First, give access to one agent only,
- 6:31not the whole crew. A developer agent
- 6:33writes codes. It probably has no reason
- 6:35to read your inbox. So, it shouldn't be
- 6:37able to. Second, start read [music]
- 6:39only. My finance dashboard, a new
- 6:41tutorial which is coming soon, reads my
- 6:42receipts straight from my Gmail. And it
- 6:44can never send an email or delete one.
- 6:46[music] It only needs to read, so
- 6:48reading is all it can do. You can always
- 6:51give more access later, but you can't
- 6:52take back an email that has already been
- 6:54sent or deleted. And the third tip, tell
- 6:56your agents that email is information,
- 6:59never instructions. Anyone can send you
- 7:02an email, and anyone can hide
- 7:03instructions inside one hoping your
- 7:05agents will follow them. This is what is
- 7:08called prompt injection. So, your agents
- 7:10reads what that email is, but never
- 7:13takes orders from it so that no one
- 7:15pulls some information from your agents
- 7:17through any tricky means. So, now you've
- 7:19got one crew connected to your tools.
- 7:21But what if you need more than one?
- 7:23Bringing us to tip number six, run
- 7:25several Hermes setups on one server. Say
- 7:28you want separate crews, one for content
- 7:30creation, one for work, one for
- 7:32personal, or one for each client you set
- 7:34agents up for. You don't need a separate
- 7:36server or system for each of them. With
- 7:39Docker, one server can run several
- 7:41completely separate Hermes setups,
- 7:43[music]
- 7:43and each one has its own isolated Hermes
- 7:46workspace, meaning its own agents, its
- 7:48own memory, and its own Telegram bot. No
- 7:51crew ever sees another crew's notes, and
- 7:54you only pay for one server. Now, at
- 7:56this [music] point, there's a lot going
- 7:58on, and I constantly keep jumping
- 8:00between Telegram folders and the
- 8:02terminal. [music]
- 8:02And that brings me to tip number seven.
- 8:04This is a big one. Build your own
- 8:06mission control. [music]
- 8:07I built one screen for all of it. I call
- 8:10it Nexura. I launch a mission and
- 8:13execute a plan set. Nothing runs until I
- 8:15approve it. And that's exactly why I'm
- 8:17comfortable letting it run. Then I watch
- 8:20the agents pass the work between
- 8:21themselves, and the results lands
- 8:23directly in docs, where I can find it
- 8:25next week instead of scrolling back
- 8:27through chats. I can chat with any agent
- 8:29from here. There's an agent city that
- 8:30lights up when an agent is working. So,
- 8:33one glance tells me who is busy. [music]
- 8:35And hey, I also set up an agent's login
- 8:37database. Whenever an agent performs a
- 8:39task, it logs in the database a summary
- 8:42of the task that they performed, the
- 8:43time that they performed that task, the
- 8:45AI model that they used, and [music] the
- 8:46agent's name. Now, this way, I can
- 8:48monitor everything that the agents
- 8:51[music] are doing. Additionally, I can
- 8:53check my machine's health and see every
- 8:55scheduled job, including the 8:00 a.m.
- 8:57brief [music] that I talked about. And
- 8:59with the voice mode, I can just talk to
- 9:01the agents, and they can also talk to me
- 9:03back if I don't feel like typing. The
- 9:04whole mission control dashboard is
- 9:06private. It runs on my own machine, and
- 9:08only my devices that I own can reach it.
- 9:11Because think about what your crew knows
- 9:13about [music] you, your emails, your
- 9:14clients, and that is why security should
- 9:16be your highest feature. [music] Now,
- 9:18before I was chatting with a bot, but
- 9:20now I'm running a crew of AI agents. You
- 9:23can build it yourself from all the
- 9:24prompts that I'm going to provide in the
- 9:25description section, or you can install
- 9:27it with an auto installer that I built
- 9:29for you guys, saving you all the AI
- 9:31tokens and time it would cost you to
- 9:32build this. Now, if you found this video
- 9:34helpful, hey, hit the thumbs up button
- 9:36and subscribe to my channel to see more
- 9:38like this. And with that, thanks for
- 9:39watching this video and I'll see you in
- 9:41the next one. Bye.
About this transcript
This page contains the full transcript of 7 Hermes Agent Upgrades You NEED To Implement Today by Komputer Mechanic, generated from the public captions YouTube serves with the video. The transcript has 2,056 words across 288 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.