# Stop Using Chatbots: This Free AI Agent Does the Work For You!!

Channel: Outskill
Video: https://www.youtube.com/watch?v=duT2ld4zWGE
Duration: 21 min
Language: English
Words: 3515
Transcript page: https://viewrankai.com/tools/youtube-transcript/duT2ld4zWGE

---

[0:00] Imagine closing your laptop at 2 p.m. and walking out while your work keeps happening without you. Your emails get sorted and answered. Your news brief lands at the [music] exact time you set. And by the time you're back for dinner, a client's website is built and live. And you were nowhere near a computer. AI can do all of that today. You just [music] haven't built it yet. Right now, the AI you're using sits in a tab and waits. [music] You open it, you ask, it

[0:26] answers, and you still do the work. Today, we're fixing that. By the end of this video, we'll build you a real AI employee. One that lives on your computer, takes orders from your phone, reads your email, drafts your replies, runs tasks on a schedule, researches the web, and ships finished work. All while you're off doing something else. For the biggest test, we gave it one client email with a five-page contract and asked it to build and deploy an entire website. What came back honestly surprised me. Stay till the end for that one. Free to start. No coding. Every step is right here on screen and set up

[1:04] takes about 10 minutes. Let's build it.

[1:14] Quick thing before we start. You've probably seen the viral AI demos going around. Someone building a landing page in a single prompt. Someone automating their whole job in a 30-second reel. It looks magical and feels impossible to actually copy. And meanwhile, chat GPT, Claude, Codex, then comes agents and automations. New tools keeps dropping every single week. Most people are more confused than empowered. That's what this Saturday's free 8-hour live AI mastermind is built around. For both tech and non-tech people. Here we will discuss the exact tools and workflows that are actually working for sales, marketing, tech, freelancing, and business. It's about research, automations, apps, workflows. Not just

[1:58] basics everyone's already teaching. The stuff you can plug into your actual work from Monday. Free for our viewers. Register from the link in the description before seats close. One quick thing before we dive in. If you're into AI builds like this, hit subscribe and join our free WhatsApp community. Every tool, prompt, and link we use in this video lives in there. So, you don't have to pause and copy anything. All linked in the description below. The tool we're using is called Hermes. It's made by a company called Nous Research. Free, open source, and

[2:30] anyone can look at exactly how it works. Which means you're not locked into one company's app. Here's the simplest way to think about it. Hermes is like a car. The AI model inside it, Claude, GPT, something free from Nvidia, whatever you want, is the engine. Hermes is the car around that engine. The controls that

[2:49] let the engine actually go somewhere. Which means one app on your computer can read your email, search the web, run tasks on a schedule, connect to other apps, and create files. And because it connects to Telegram, Slack, Discord, and WhatsApp, you can message it from your phone, and your computer does the work. So, let's install it first. Head to hermes-agent.nusresearch.com and click download. Windows and Mac get a desktop app. Linux installs with one terminal command. Open the installer and click install. It handles everything itself. Nine steps, no decisions. Just let it run. When it's done, click

[3:28] launch. Now, Hermes needs a brain. Remember the car analogy? The AI model is the engine. We need to drop one in before Hermes can go anywhere. For this video, we're using a free model from Nvidia. It's good, it costs nothing, and the only limit is about 40 requests per minute. Plenty for everyday use. Click

[3:47] other providers, then open router. Search for Nvidia and click it. Hermes asks for an API key. An API key is basically a password that lets Hermes talk to Nvidia on your behalf. Open build.nvidia.com. Sign up or log in. Click your profile, then generate API key. Give it a name, hit copy, save it somewhere safe. Nvidia only shows you this key once. Back in

[4:12] Hermes, paste the key, click connect. Now pick a model. You'll see options like Nemotron 3 Ultra 550B, GLM 5.2, Kimmy. We're using Nvidia Nemotron 3 Super 120B. Strong, fast, free. Click begin. Let's test it. That's it. Your AI employee has a brain. Quick note. If you'd rather use Claude, GPT, or any paid model, you can swap the engine anytime from the models menu. For professional work, we'd pick Claude from Anthropic. Click models, then edit models, add provider, and connect another provider. Click open router, search for Anthropic, and paste a key

[4:52] from platform.claude.com. Just know it runs on API credits, not your Claude subscription. Max plans include three monthly credits. On Pro, you'll need to add at least $5. We're sticking with Nvidia because it's free. Now when you restart the application, Hermes asks for your name, an accent color, and a layout. Pick basic. That's the everyday one. Elite is for developers with a terminal. Skip the apps check, the plugins check, and the tour. Here's what you're looking at. On the left, new session, capabilities, messaging. Open capabilities. Four tabs matter here. Skills are small packages that teach Hermes one specific job, and installing one is super easy. Just click

[5:36] on the skill you want, like Whisper, which turns speech into text and supports 99 languages. One click and it's installed on your PC. Tools add stuff like image or video generation. Those need separate paid keys. Connectors hook Hermes into other apps like Telegram, WhatsApp, Discord, and others. Remember this one. We'll use it later to put the website online. And finally, plugins give Hermes additional

[6:01] powers like memory and web browsing. There's an entire catalog with 742 plugins across nine categories and you can even submit your own. So, Hermes is running, but right now it only works when you're sitting at your computer. Let's fix that. Picture this, you're out for lunch and need to send an email. Instead of opening your laptop, just message your computer from your phone and Hermes handles it. One catch, your computer needs to stay on with Hermes running. That's it. For messaging, we're using Telegram. It's free, it's on every

[6:35] phone, and it's the simplest to set up. In Hermes, click messaging, then Telegram. There's a quick QR setup, but it doesn't always work, so we're doing it manually. Takes 2 minutes. You only need three things, a bot, a token, and your user ID. First, the bot. Open Telegram on your phone, search for @BotFather, hit start, send /newbot. It asks for a name. We're calling ours Outskill Agent, then a username. No spaces, must end in bot. Something like outskill_personal_agent_bot works. Second, the token. BotFather sends back a token. This is your bot's password. Keep it private. If it ever leaks, send /revoke and you'll get a new one. Copy the token, paste it into

[7:22] Hermes under bot token. Third, your ID. Hermes needs your Telegram user ID, so only you can message your bot. Without this, Hermes won't respond to anyone including you. Search for @userinfobot in Telegram and tap start. It sends your ID. Paste that into Hermes, click save changes. Hermes restarts the connection and honestly, ours got stuck. We tapped start on our bot and got an error. Don't quit there. Here's a trick worth remembering. Anytime you hit an error in Hermes, open Claude or a free tool like Google Anti-Gravity. Paste the error or

[7:58] attach a screenshot and ask what to do. It walks you through the fix in plain English. We did exactly that. Claude spotted the problem, gave us the fix, and the bot came back to life. Remember this move. You're going to need it again. Now, the moment of truth. Back in Telegram, tap your bot's link, say, "How are you?" That's it. You just texted your computer and it answered. From a lunch table, a train, another country, anywhere with signal. Slack, Discord, and WhatsApp work the same way. Each one

[8:29] has a setup guide button inside Hermes. Remember that client email from lunch? Right now, Hermes can't even see your inbox. Let's change that. To connect email, we're going back to Claude. Same conversation we used earlier. So, Claude already knows we're working on Hermes. Ask, "I need to connect my email to Hermes. Can you help?" Claude gives two options. Option A is Himalaya, fast setup, email only. Option B is a Google Workspace login, slower, but adds calendar and drive on top. We just need

[9:00] email, so we're going with Himalaya. Himalaya uses a 16-character app password. It sounds technical, but it's just a few clicks in your Google account. An app password is a special password Google makes for one specific app, not your regular Gmail password. It works with a personal Google account. Work and school accounts usually can't make one. First, turn on two-step verification. Google won't let you create an app password without it. Then, click the app passwords link Claude gave you. Pick your account, name the

[9:31] password, copy the 16 characters. Important security rule. Never paste passwords or API keys into any AI chat, including Claude. Instead, ask, "I have the 16-character password. Where do I paste it?" Claude will tell you where it actually goes. For us, that was the terminal, not the chat. Let's test it. On Telegram, send, "What are my five

[9:54] latest unread emails?" And error. Remember that move from earlier? Here's where it pays off. Claude's fix was simple. Send {slash} new in Telegram. That starts a fresh chat, so Hermes picks up the email skill it just learned. Then, use the Himalaya skill to show my five latest unread emails. And there they are. Your five latest unread emails right inside Telegram. Your inbox checked from anywhere without opening Gmail. One safety tip from us, give Hermes its own Gmail account, not your main inbox. If anything ever goes wrong, your personal mail stays safe. But typing every request on a phone keyboard gets old. What if you could just talk to

[10:37] Hermes? One voice note to sort days of email. Everything else done with just your voice. Hermes understands voice both on your phone and on your computer. Let's set up both. And stay for the last one because you won't even need to press a button. Back in Claude, prompt set up voice for my Telegram bot. Claude installs two free tools on your computer. Whisper turns your voice into text. Piper turns Hermes's replies into speech. Both run locally. No cloud, no monthly fee. Now in Telegram, send {slash} voice on so Hermes can answer you with voice, too. Now press and hold the mic button and speak. Our first try

[11:16] threw an error. You know the move by now. Paste it into Claude, apply the fix, try again. Let's give it a real task. Press, hold, and say, "Check my emails from the last 10 days. Tell me which ones need a reply and draft a reply for each." Watch what happens. Hermes converts the voice note to text, checks the inbox with Himalaya, picks out emails that need responses, drafts a

[11:39] reply for each one. One voice note, 10 days of email sorted without opening a single message. Hermes can send replies for you, too. But here's our rule, always ask it to show the draft first, review it, then say send it, and the reply goes to the actual person, not back to you. For voice in another language, just ask Claude to switch Whisper to that language. Now, same thing on your computer. No typing there, either. In the Hermes app, click start voice conversation. It's the round button next to the microphone. Allow mic access the first time. Note, you need to toggle on speech-to-text in tools. Just talk. When

[12:18] you pause, Hermes types what you said, sends it, and answers out loud. You can say stop to end. There's one more trick, and it's my favorite, a wake word. Type {slash} wake on in Hermes. Now you can say, "Hey Hermes." from anywhere in the room, and it starts a fresh chat and listens. Quick privacy note, the wake

[12:38] word listening is 100% on your computer. No audio ever leaves your machine until you actually give it a command. Even then, only the text of your request goes to the model. If this is already useful, hit subscribe. The next part is where it gets wild. What if Hermes could run tasks for you before you even get out of bed? This is where Hermes starts working without you. It can run jobs on a schedule. Think of them as alarm clocks for your AI. You choose the task and the time, and Hermes runs it automatically and sends you the result. Let's set one up. A morning brief with the latest AI

[13:13] news and stock market updates. Here's the exact prompt we sent to Hermes on Telegram. Create a scheduled job called morning brief. Run it every day at 7:00 a.m. Send me the latest AI news and stock market updates with one key takeaway. Quick disclaimer, that's news, not financial advice. We use Telegram on purpose. Hermes replies in the same chat, so the brief lands right on your phone. Hermes shows the job and asks if you want to proceed. Reply one to

[13:41] confirm. Now let's make sure it's saved. Open the terminal, type Hermes dashboard, and press enter. The dashboard opens in your browser. Click cron. There's your morning brief. You can check or edit it here anytime. One catch, your computer needs to be on, awake, and online with Hermes running. If your laptop is asleep at 7:00 a.m., the brief arrives late. It runs as soon as your computer is back on. It's not a cloud server running 24/7. Don't want to wait till tomorrow? Just message Hermes on Telegram. Run my morning brief job now. And there it is. Today's AI headlines, the market moves, and one clear takeaway in one place before your

[14:21] first coffee. No more jumping between websites. So, Hermes works from your phone, reads your email, understands voice, and runs tasks on a schedule. But, setting it up is one thing. Seeing it handle a real client project is where things get interesting. Let's do that. This is the one we promised at the start. One client email turned into a working website live on the internet 3 days before deadline. First, send Hermes this. Check today's emails and draft replies to anything that needs a reply today. Only one email comes back needing a reply, and it's a big one. The client, Nossa Familia Coffee Company, a roastery with two cafes in Portland, wants a

[15:00] brand new website. $4,800 deal. Work starts tomorrow. First preview due in 5 days. Hermes has already drafted a reply. We read it, said, "Go ahead and reply now." Sent. Then, now summarize the content of the PDF. Five pages of contract broken down in plain English in about a minute. Required pages, brand colors, fonts, menu, prices, deadlines, payments, even the small print. Now, here's where most people mess up with AI website builders. They type, "Build me a

[15:32] website." and get something generic. Good results need two things: the right tools and clear instructions. First, the tools. Create a new folder on the desktop named after the coffee shop. Then, download the Impeccable GitHub skill. Impeccable is a free design skill. It helps the AI make better decisions about typography, colors, spacing, layout, the stuff that makes a

[15:55] site feel designed, not auto-generated. Hermes installs it and asks if it should load the skill. Approve. Now, the actual brief. Load the Impeccable skill. Read the client's PDF. Build the website exactly as it describes. Check your work against their requirements. Send me screenshots. Do not email the client until I approve. Look at what we just told it. Which tool to use, what to read, what to build, how to check its work, and what not to do. That's how you brief an AI properly. Hermes doesn't rush into code. It plans first. It proposes a warm oat background, a deep blue harbor header, the client's fonts, and a live board showing which cafe is

[16:38] currently open in Portland time. It also catches that one of the brand colors is too light to read as body text and suggests a fix. Nobody asked it to. Approve the plan. Half an hour later, version one runs on localhost. That's your own computer, not online yet. It works, but it doesn't look professional yet. So, we ask for a pro redesign with the free taste skill built to kill the generic AI look. And this honestly surprised us. Hermes notices that Nordhavn is Danish for North Harbor. So, the site opens on a painting of Copenhagen's harbor in 1900, fading into Portland's waterfront in 1898. Nordic roots, new home. Reviews appear as torn

[17:20] paper notes. Then it tests on phones, tablets, and laptops. 96 out of 100 for mobile performance, a perfect 100 for accessibility. First email to finish site, about an hour and a half. Work that would normally need a designer and a developer. But there's one problem. The site is still on our computer. The client can't see it. Let's put it online. We're using Vercel. It's free for personal projects, $20 a month on pro for paid client work. Remember the connector section from earlier? This is where it pays off in capabilities. Open connectors. Search Vercel. Click connect. Create an account if you don't have one and authorize Hermes. Now, one

[18:03] instruction. Verify Vercel is connected, fix any issues, and deploy the website on Vercel. Deploy just means put the site online so anyone with the link can see it. Halfway through, Hermes asks for one more approval. We review the action, approve. And about 4 minutes later, the site is live. In our run, Hermes also quietly keeps the site out of Google search while it's a preview. So, a random search for the client doesn't find a half-finished version. Now, the client needs the preview. Just write, "Draft a reply to the client with the

[18:36] live link and the project progress. Professional, formal tone. Don't send until I say so." An MVP, by the way, is just a first working version of a product. Hermes writes a progress update, drops in the live link, attaches the preview file. Exactly what the contract asked for. "Show me the draft." We read it. "Deliver now." Sent. The preview was due October 20th. It went out October 17th, 3 days early. And Hermes points out one more thing. With the preview delivered, the next contract payment of $1,920 can now be invoiced. Think about what just happened. One email came in. Hermes reviewed the request, summarized the contract, drafted the reply, built the

[19:21] website, tested it on multiple devices, deployed it online, and prepared the client update. You approved. Hermes did the work. Is it perfect? Not always. AI-built work can still have rough edges, even security bugs. So, always review it before the final launch. But, this is how professionals work today. AI does the heavy lifting across multiple projects at once, and you steer it from anywhere in the world. Before you hand Hermes your whole inbox, four things to know. One, your computer needs to stay on, awake, and online for local tasks and scheduled jobs to actually run. On a Windows laptop, keep it plugged in and set it to do nothing when the lid

[20:04] closes. Then, it keeps working even with the lid shut. Two, free AI models have usage limits. Paid models, Claude, GPT, cost money through API usage. Nothing is unlimited. Three, Vercel's free plan is for personal use only. For paid client work, you need Vercel's Pro plan, about $20 a month. Four, expect errors during setup. We hit several. Claude helped us fix everyone. Troubleshooting is part of the process, not a sign you're doing something wrong. And always review drafts, check the files Hermes creates,

[20:40] and approve sensitive actions yourself. You're the director. Hermes is the team. This used to mean switching between five apps and doing every step yourself. Now, Hermes brings it all into one place. You're still the director. You decide, you review, you approve. Hermes handles the repetitive work. Everything we used in this video, the prompts, the design skills, the connectors, lives in our free WhatsApp community. Grab it from the link in the description. If you

[21:08] haven't joined yet, now's a good time. And tell us in the comments, what's the first job you'd give your own AI employee? The best ideas usually become the next video. Watch the next video on screen. See you there.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt