13:38Alex Finn · Tutorial
The new Hermes Agent update has me speechless
Eight features in Hermes Agent 0.17 that move it from chatbot to ambient operating system.
June 24thThe argument in one line.
Hermes Agent's eight new features—including session recall, background multitasking, computer control, and auto task breakdown—position it as superior to existing AI agents for autonomous work delegation.
Read if. Skip if.
READ IF YOU ARE…
- An existing Hermes Agent user who has not updated recently and wants a fast rundown of 8 major new features — session recall, background tasks, Grok OAuth, and Codex CLI — before deciding which to enable.
- Someone frustrated that their agent can only do one thing at a time and wants to understand how background tasks let a single instance handle multiple concurrent research jobs.
- A power user tracking Hermes vs OpenClaw parity who wants to see where the gap has widened, specifically around memory recall and multi-task execution.
- A creator or analyst who wants live examples of the new tweet search and Kanban task-generation features applied to real content and research workflows.
SKIP IF…
- You have not yet installed Hermes Agent — this is an update overview for existing users, not a setup guide.
- You are looking for deep configuration instructions; each feature is demonstrated in a 1-2 minute live demo with no deep setup walkthrough.
The full version, fast.
Hermes Agent's latest release closes the gap with OpenClaw by turning a single AI assistant into a true multitasking autonomous worker. The update layers eight capabilities onto the core agent: session recall pulls past conversations by date without burning tokens, /background queues parallel jobs you can keep chatting through, Grok 4.3 OAuth adds real-time tweet search and native text-to-video generation, native Codex CLI lets the orchestrator delegate coding to a cheaper specialist, and computer use drives your desktop while you're away. The Kanban auto-task generator is the through-line: drop a goal into triage and Hermes shards it into subtasks routed to sub-agents. Treat morning to-do dumps as triage input and the fleet works while you make coffee.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:29
01 · Cold open + promise
Mind-blown hook, promises 8 changes + autonomous employee payoff.
00:29 – 02:11
02 · Feature 1: Session recall memory
Token-free programmatic session recall. Live demo: query what was worked on May 10.
02:11 – 04:01
03 · Feature 2: Background tasks
/background command enables true multitasking. 3 parallel research tasks + live chat simultaneously.
04:01 – 05:27
04 · Features 3+4: Grok OAuth + real-time tweet search
Dual-use X subscription inside Hermes. Enables live tweet search for content research.
05:27 – 07:00
05 · Feature 5: Native Codex CLI
Spawn Claude Code as a worker sub-agent from Hermes. Demo: 3D FPS in Three.js as a background task.
07:00 – 09:03
06 · Feature 6: Computer use
Agent sees and controls the desktop. Demo: read Notion calendar + add event (15-min timing miss).
09:03 – 10:43
07 · Feature 7: Native video generation
Text-to-video via Grok Imagine inside Telegram. Demo: dragon vs horse AI video.
10:43 – 13:13
08 · Feature 8: Auto Kanban task generation
Drop a goal in triage; Hermes breaks into subtasks and assigns to sub-agents automatically.
13:13 – 13:43
09 · CTA
Like + subscribe, Vibe Coding Academy Skool plug, comment-bait asking what video to make next.
Lines worth screenshotting.
- Session recall memory lets Hermes remember every session by date without consuming AI tokens — the recall is programmatic, not LLM-driven.
- Background tasks allow multiple parallel work items without spinning up separate agents — you can still communicate with Hermes while it grinds through multiple background jobs simultaneously.
- The /background slash command assigns a task to background execution while keeping the main conversation thread open for new requests.
- Grok OAuth lets users double-use their existing X subscription as an AI model provider inside Hermes — no separate API key required.
- Grok OAuth integration enables real-time tweet and post search inside Hermes — useful for content research, trend spotting, and competitive monitoring.
- Native Codex CLI integration lets Hermes spin up a Codex worker for vibe coding tasks, which saves money when using ChatGPT as the orchestrator because coding uses the optimized model.
- Computer use lets Hermes see your screen, click, and complete any task you can do manually — including reading your Notion calendar and adding new events.
- Video generation is now available inside Hermes — you can request a video from a prompt and receive a generated clip directly through the agent.
- Auto Kanban task assignment means Hermes can read a project and automatically populate a task board with structured issues — closing the loop between agent sessions and project tracking.
- Combining /background with native Codex CLI means Hermes can vibe code an entire application in the background while you work on something else in the foreground.
- Asking 'what were we working on on May 10?' and getting a detailed session summary is the user-visible change from session recall — it removes the 'memory problem' complaint from every previous AI agent.
- Updating Hermes is a single command: 'Hermes, update to the latest version' — the self-update mechanism requires no manual file replacement.
The numbered-list demo formula.
Steal this format
Eight features, one slide that grows, screen recordings that prove every claim — this is the lowest-effort high-trust format for AI tool update videos.
- Open with the number: '8 changes that totally changed the way I use X' sets expectation and keeps viewers through the end.
- Build your list on screen incrementally — each new item appearing is a visual progress bar signaling more is coming.
- Demo every claim live — even an imperfect demo (calendar event landed 15 min early) is more credible than a screenshot.
- Combine related features into one demo to show two items with one screen recording (background tasks + Codex CLI).
- Use the morning routine as your lifestyle CTA: 'by the time I come back from coffee' is a concrete image anyone can steal.
- End with a three-option comment-bait question — gives viewers a reason to engage and tells you what to make next.
Terms worth knowing.
- Hermes Agent
- An AI agent platform that runs autonomous tasks across tools like Telegram, with features for memory, background jobs, and multi-agent orchestration.
- Session recall
- A memory feature that lets an AI agent retrieve everything discussed in past sessions by date or topic without consuming AI tokens, using programmatic lookup instead of model calls.
- AI tokens
- The billable units of text that large language models read and generate; most agent features cost money in proportion to how many tokens they consume.
- Background tasks
- A mode where an agent runs multiple jobs concurrently behind the scenes while the user keeps chatting with it, avoiding the need to spin up separate agent instances.
- Multi-agent approach
- An architecture where several AI agents run in parallel and divide work between them, typically requiring more compute and token spend than a single-agent setup.
- OAuth
- An authorization standard that lets an app act on your behalf in another service using your existing login, so you can reuse a subscription instead of paying separately for API access.
- Grok 4.3
- A large language model from xAI, available through X subscriptions, used here as a provider that can also search real-time posts on X.
- Orchestrator model
- The primary AI model an agent uses to plan and coordinate work, which can delegate subtasks to other specialized models or tools.
- Codex CLI
- OpenAI's command-line coding agent that runs locally to read, write, and edit code in a project folder, often used as a worker process inside larger agent systems.
- Vibe coding
- Building software primarily by describing intent to an AI agent in natural language and letting it generate the code, instead of writing it line by line.
- Opus
- Anthropic's top-tier Claude model, capable but expensive per token, often used as an orchestrator rather than for bulk code generation.
- Computer use
- An agent capability that lets the AI see the screen and control the mouse and keyboard to operate desktop apps the way a human would.
- Text-to-video
- AI generation that turns a written prompt into a short video clip, typically a few seconds long.
- Grok Imagine
- xAI's image and video generation tool, accessible through Grok, used here to produce short AI videos directly inside chat.
- Higgsfield
- An AI video generation service heavily promoted in YouTube sponsorships, referenced here as a third-party alternative the user no longer needs.
- Kanban board
- A visual task management layout with columns like Triage, To Do, and Done, where work items move across columns as they progress.
- Triage
- An intake column on a Kanban board where new tasks land before being refined, broken down, or assigned to someone to work on.
- Sub agents
- Specialized AI workers spawned by a main agent to handle individual subtasks in parallel, each scoped to one piece of a larger job.
- Three.js
- A JavaScript library for rendering 3D graphics in the browser using WebGL, commonly used to build games and interactive scenes in a single HTML file.
Things they pointed at.
00:00productHermes Agent
04:01productGrok 4.3
05:27productCodex CLI
Lines you could clip.
12:42
“By the time I come back, I have like 40 different subtasks that all my agents are just working on.”
00:07
“Hermes agent just had its biggest update yet, and my mind is blown.”
02:02
“This takes it to the next level.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphorstory
00:00Hermes agent just had its biggest update yet, and my mind is blown. This is the update that I think officially puts it ahead of Open Claw.
00:09In this video, I'm going to cover the eight changes that have totally changed the way I use Hermes agent. If you stick with me until the end, you are going to have your own doing more work for you than you can ever expect in your entire life.
00:26Now let's lock in and get into it. So I don't do a ton of update videos, but this latest update was so good that I wanted to make a video around it. Let's go through these eight changes.
00:36The first update is dramatically improved memory.
00:40They added into Hermes agent what is called session recall. Now your agent can recall every single thing you've done and talked about by session. So you can do things like say, hey.
00:52What did we do or talk about last Thursday? Or, hey. What did we talk about two months ago?
00:56And it will be able to recall every single thing from every single session without having to use token calls at all. They implemented a very cool technology that doesn't rely on AI tokens to recall memory.
01:09It's just able to do it programmatically. Let me show you how this works. So if I'm coming in here and I wanna know what we did so if I'm coming in here and I wanna know what we did, like, between Monday and Wednesday last week.
01:20Right? Like, I'm working on a couple projects. I wanna see what we discussed last week.
01:25And by the way, again, you can do this. Hey. What were we discussing, like, two and a half months ago on this specific day at this time?
01:31But watch this. What were we working on on May 10? And boom.
01:36Here we go. Now it's going to go, and it's going to look at what we were doing May 10. Okay.
01:42On May 10, we did a deep equity research project together. You sent me a tweet with 40 tickers across 10 themes. I fanned the work.
01:48It told me look. It looks step by step. It told me exactly what we did and what I sent it.
01:53This improves the memory so much. One of the biggest complaints I get about OpenClaw and really every other AI agent out there is the memory. This takes it to the next level.
02:03So make sure you update your Hermes. All you gotta do is say, Hermes, update to the latest version.
02:11Hit enter. It'll update there, and you will get this new session recall in your memory, which will make it so much better at remembering everything you've done in the past. The next major feature I think is awesome is background task.
02:23Background tasks allow you to give your agent multiple tasks to do that it will work on in the background. One of the issues with these AI agents is when you give it a task, it's kind of busy now. You can't have it do other things.
02:36It's very hard to multitask unless you set up a multi agent approach. But to do a multi agent approach, you either need to spend a ton of money on your AI tokens, or you need to have a really good computer with a lot of agents running, not with background tasks.
02:51Now your agent can do multiple things at once without having to spin up a ton of agents. Watch this. So to give your agent background tasks, all you need to do is do slash background and then give it a task.
03:03So for instance, watch this. I'm gonna have a go off and research the top 10 AI agent startups from the last thirty days. So now it's getting to work.
03:12I'm gonna have it research my last 50 newsletters I sent out. Let me know which ones are working.
03:18And I'm gonna have it research the top trending YouTube videos in AI. Boom. And while it's doing all that, I'm gonna say, what time is the Spurs game tonight?
03:28And I'm gonna hit enter, and it's gonna be able to communicate with me even though it's doing all these background tasks at once. So this is allowing it to really do true multitasking without having to set up 20 different agents.
03:41And as you can see here, told me the time of the game. It's actually starting one hour from now, so I gotta finish this filming quickly. And then it went in and it found all these trending videos on Claude Code, Open Claw, Hermes agent.
03:52It did all those things at the same time while it's still working on the two other background tasks I gave it. This is the best way so far to do multitasking with your agent slash background.
04:03Give it as many tasks as you want, and you'll still be able to talk and communicate with your agent, which is amazing. The next two are closely related, and this is pretty cool. We'll go through this quickly.
04:12Grok 4.3 OAuth. So Grok 4.3 just released in full.
04:17It's a solid model. It gets the job done. I think the advantage here is twofold.
04:22One is a lot of people are subscribed to Grok 4.3 because they're subscribed on x. So now you can double use that OAuth inside your Hermes agent if you wanna save a little money. Next update for Hermes agent, which is the fact that you can use Grok OAuth to search real time tweets.
04:40So this is awesome. If you use your Grok 4.3 OAuth, you can have it search x for posts in real time.
04:48So find trending news, find trending topics, find trending content. It's really, really cool. All you need to do is add Grok 4.3 as a provider.
04:57So even if you're using Claude or Chad GPT as your orchestrator model, you can add in Grok 4.3 and say, hey.
05:05Anytime I ask you about posts, use Grok 4.3, and it will be able to use it as a muscle. So if you do a lot of research around content, which is something I do quite often, make sure to plug in your Grok 4.3 OAuth so you can do a lot better native research on tweets.
05:21The next awesome update in the Hermes agent is native codec CLI use.
05:28Now your Hermes agent can natively built into Hermes, use the codec CLI. So even if you're on ChadGBT 5.5 or you are on Opus as your orchestrator, it can spin up as a worker a Codec CLI session, which basically means it can do its own vibe coding, which if you have subscription, will save you tons of money because now you don't have to have your Opus do a whole ton of coding for you, which will be really expensive.
05:55Or if you're already using ChadGBT as an orchestrator model, you'll get better results because it will use its native coding functionality.
06:03So all you need to do is go in, and you can even do this as a background task. So let's combine a couple of these things here and say, use codecs to build a three d first person shooter using three JS in a single HTML file.
06:19And I'm gonna hit enter on that. And now the background task is going. It's actually using the native codec CLI, which in my opinion is the best vibe coding tool out there right now to build this out.
06:31So now I can go have Hermes do other things, research other things for me. And in the background, Codec CLI is building out an entire application for us. You need to be using Codec CLI, especially if you do any sort of vibe coding with your AI agents.
06:46So really, really cool. It builds its own folder inside the Hermes folder where that project will live, and it's spinning up codec.
06:54So it's using a separate agent, your codec CLI to actually build it out, which is really cool. The next update I really love, number six here, is computer use.
07:04Your agent now can control your computer. Anything you can do on your computer, your agent can now do as well. It can see what you're doing.
07:13It can complete tasks. It can click around. Anything you want, it can work alongside with you.
07:18This is really cool. So for instance, if I wanna go in, I wanna say, look at my Notion calendar and let me know which events I have today and then add a new event for 7PM, and I hit enter on that, it is going to now let me pull open my calendar.
07:40It is going to let's look at this side by side here. It's using computer use as you can see here. It should see a couple different meetings here, meeting with Ben, meeting with Angela, meeting with Allison.
07:51And let's see here. There there it is. Seize the events.
07:55Note your 7PM slot overlaps the IO Gemini. Go on to Google IO tomorrow. Let's say, make an event called film video.
08:05I'm gonna hit enter on that. And now we should be able to watch it actually build an event on my calendar using my computer.
08:15So I'm not touching my computer at all right now. It is going in, and it is actually looks like it's going to add the event. So, boom, you see a new event.
08:24I haven't moved my hands once. It added the event to the calendar. It looks like there's one issue.
08:30It did hit for 06:45. So it did fifteen minutes early. So maybe telling it to add events might not be the best use case here.
08:37Point being is your agent can now control your computer and do whatever you need for you. One way I think this could really be helpful is if you're out and about on the road wherever and your computer's still running, you go on your phone, you go on Telegram on your phone and say, oh, man.
08:52I forgot I had this open. Can you do this for me real quick? Can you change this around for me real quick?
08:56Could you exit out of this for me real quick? And it'll be able to actually control your computer and do that for you, which is really, really cool. The next really awesome feature, we got two more, and these both are amazing, is native video generation.
09:11Now your Hermes agent can go text to video or even photo to video natively right through your chat. So if you say something like, generate me a video of a dragon, it'll be able to generate you an AI video.
09:23You don't have to go to other websites or other experiences to do it. As long as you have this hooked in to your Grok OAuth. So, again, this is another great reason to have Grok OAuth hooked in.
09:35You can use it to generate videos with Grok and Magic. So let's do this. Generate me a short video of a dragon fighting a horse.
09:45We're gonna hit enter on that. And now you can see it's using its video generate tool. It wrote its own prompt to do this.
09:50It's gonna generate us a video straight in Telegram here. No need to go to any other AI tool. No need to listen to the Higgs field chills all over YouTube having you go to that scam alleged website.
10:02It all happens native inside of Telegram. Look at this.
10:06Here you go. Red dragon versus a white horse. That's all done in Grok Imagine.
10:11All done right in Telegram. There you go. There's your AI generated video.
10:16Really cool. I mean, are you gonna ever need videos of dragons versus horses? No.
10:20But the fact that you can generate these videos natively without having to go to other websites is incredible. So you just set up your Grok OAuth.
10:29By the way, all you need to do that is in your terminal. Do Hermes tools. Hit enter in the terminal.
10:34Then you can choose the video generation tool, enable it, and then it'll just have you sign in to Grok, and you're good to go.
10:41Really, really easy to do. And the last one is auto Kanban task generation.
10:48This is a really cool one. Stick around for this. Now you can drop goals into your Hermes Kanban board, and it will auto turn that goal into a bunch of tasks and assign those tasks to sub agents.
11:01This is really cool. Watch this. So this is amazing.
11:04For anyone who hasn't played around with it yet, make sure you go in here. Go to your Hermes dashboard. You can just go in a terminal, say Hermes dashboard.
11:13Then you click on Kanban. Once in Kanban, you can start adding tasks to your triage.
11:19And what's going to happen is whatever tasks you add to your triage, Hermes will automatically take that task, break it into subtasks, and then assign those subtasks to all your different agents.
11:32So if you have multiple agents, multiple sub agents, they'll get assigned, and they'll get to work on your task. Watch this.
11:39So I'm gonna go into triage. I'm gonna say new task. I'm gonna say script me a video for a Hermes master class, then design a thumbnail.
11:51And I'm going to let's see. That's all I'm gonna say, use the Garak Imagine for that. I'm gonna say create.
11:58Now the task is in triage. Watch what happens in a second here. And boom.
12:02Look at this. Automatically in To do went script me a video for Hermes agent. Another one went write full master class video script.
12:10Another one one did generate Hermes master class thumbnail. So it took my task. It broke it into subtasks, and now they're all in To Do.
12:20And as you can see here, each are assigned to a different sub agent. So each are gonna get that task now. This is amazing.
12:27It's gonna get to work. And this is great because if you have big lofty tasks, all you need to do is go in, go into your Kanban board, add them to triage.
12:38And if you have a list of 10 of them, you can just put them in, and Hermes will go and just break them all down into their little components and build them out for you. My favorite way to use this is every morning I wake up. I get my to do list.
12:50I write this little to do list down here every single morning. I take those tasks. I put it in a triage.
12:56I hit enter. I go make my coffee. By the time I come back, I have, like, 40 different subtasks that all my agents are just working on.
13:04Saves me so much time. It is awesome. These were the big updates.
13:08If you learned anything at all, make sure to leave a like down below. Subscribe. Turn on notifications.
13:14Also, join the Vibe Coding Academy. I I do a live boot camp on Hermes and other AI tools every single week. You can join in, ask me questions, get tech support, whatever you want.
13:23Hit the link down below if that's the number one AI community on planet Earth. Let me know in the comments what video do you want from me next. Full Hermes use cases, codex masterclass, Claude code masterclass.
13:33Super curious. Let me know in the comments. Hope this was helpful.
13:36I'm so grateful you'd stick with me here. Watch all these videos. Hope they're really helpful.
13:41I'll see you in the next video.
The Hook
The bait, then the rug-pull.
Alex Finn opens with a declaration — Hermes Agent just had its biggest update yet, and he is blown. Eight features, one video, a 24/7 autonomous employee waiting at the end. The numbered list on screen starts at one and does not stop until it hits eight.
Named ideas worth stealing.
00:40concept
Session Recall
Token-free programmatic memory that recalls every past session by date without consuming AI tokens.
Steal forPitch angle for any tool with long-term memory: no token cost is a concrete differentiator worth calling out explicitly.
02:51concept
/background command pattern
Delegate multiple long-running tasks to background workers while keeping main chat available for synchronous queries.
Steal forFramework for explaining async agent workflows to non-technical audiences: give it tasks, go make coffee, come back to 40 subtasks done.
05:27model
Orchestrator + Worker model
Use Claude/ChatGPT as orchestrator, plug in Grok/Codex as specialist workers for specific tasks.
Steal forApplicable to any multi-model AI workflow explanation — the muscle framing is reusable.
12:40concept
Morning Kanban routine
- Wake up
- Write to-do list
- Drop into Kanban triage
- Go make coffee
- Return to 40 subtasks in progress
Use Kanban triage as a daily delegation ritual — agent auto-decomposes and assigns while you step away.
Steal forStrong lifestyle-angle hook for any automation video: by the time I come back from coffee, 40 tasks are already running.
How they asked for the click.
VERBAL ASK
13:13next-video
“Let me know in the comments what video do you want from me next.”
Three-option comment bait (Hermes use cases / Codex masterclass / Claude Code masterclass) plus subscribe ask and Skool community link. Clean sandwich execution.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
