I Built Jarvis in One Afternoon with This Free AI Tool
60sThe promise of building a personal AI assistant like Jarvis for free is highly appealing and taps into the fascination with AI and productivity.
▶ Play Clip
"Delivers on the promise of building a Jarvis-like assistant with a free tool, though the 'INSANE' hype is slightly oversold."
This video demonstrates how to build a personal AI assistant crew called 'Jarvis' using Open Mousebot, a free, open-source alternative to Grockbot. The creator walks through a seven-step setup process, then showcases the crew handling four real business tasks: sorting a support inbox, researching a sales lead, planning a founder's week, and generating a week of content. The video emphasizes the importance of giving each bot clear instructions and using the 'ask me first' setting to maintain human control.
The creator introduces Jarvis, an AI chief of staff built in one afternoon using Open Mousebot, a free, open-source alternative to Grockbot. Jarvis coordinates a crew of AI bots that can read websites, write in the user's voice, create 3D scenes, and manage the inbox.
Open Mousebot is a chat app where each contact is an AI agent with its own job, personality, and brain. It's open-source, free, and runs locally. It integrates with existing AI tools like Claude, Codeex, and Grock, and supports multiple models including Grock, Deepseek, and Muse. It has over 23,000 GitHub stars and runs on Mac, Windows, and Ubuntu.
Download Open Mousebot from GitHub, open it like any other app, and on Mac, drag it into Applications. The Mac version is signed by Apple.
Click the plus icon in the sidebar, select 'new bot', and rename it to 'Jarvis'.
The most important step: open bot settings and write a job description. The creator's instructions for Jarvis include: 'You are Jarvis, my calm chief of staff with a light British manner. Short, warm sentences. Open with right away or done when it fits. Never do the whole job yourself. Brief scout, quill, pixel, and relay in parallel. Check their work and hand back one tidy summary with the files. Drafts only. Never send or post.' Also, set the 'job' field to 'chief of staff runs the crew'.
In voice settings, choose a provider (11 Labs, Fish Audio, Grock, Gemini, OpenAI, or built-in Mac voices). The creator uses the built-in British voice 'Daniel' and enables 'read replies aloud'.
Use the templates page to add bots: Scout (researcher), Quill (writer), Pixel (designer), and Relay (inbox manager). Replace template instructions with a short paragraph describing each bot's role.
Click the model name at the top of a chat to open the picker. Assign different models to each bot: Jarvis on Grock 4.7, Scout and Pixel on DeepSeek V4 Pro, Quill on Muse Wonder, and Relay on DeepSeek V4 Flash. This allows switching models if one gets stuck.
Create a new group chat called 'Jarvis HQ', add all five bots, and set Jarvis as the lead. Add group instructions: 'Jarvis leads. He splits every job across the crew in parallel. Drafts only. Save every file in this folder and Pixel finishes every job with a 3D scene.'
A software founder's support inbox with 8 emails. Jarvis sorts them into urgent/today/later, identifies the recurring export bug, drafts replies, and creates a 3D scene. Relay flags that the bug was only reported by two customers, so Scout confirms before calling it a known issue. The crew finishes in ~18 minutes.
For an SEO freelancer, Scout opens a real website (openmousebot site) and finds three fixes: inconsistent product naming, unclear main heading, and page title leading with another product's name. Quill writes a call brief, Relay drafts a follow-up email, and Jarvis catches a false line in the email.
For an overwhelmed founder, Jarvis reads notes and picks top three priorities: export bug threads, partner demo slides, and investor update. It suggests dropping hiring a designer and delegating the welcome email. Relay marks handoff notes as drafts and clarifies which words are its own.
From one idea ('Most people use AI like a search box, but the people getting ahead use it like a team'), the crew generates a LinkedIn post, carousel, video script, thread, and newsletter intro, all telling one story across the week. Everything is saved as drafts.
This setting controls whether bots act autonomously or ask for permission. For drafts, allow 'just do it'; for anything leaving the business, use 'ask me first'. This ensures human control.
The old way: you do the work with AI helping, checking everything yourself. The new way: the crew does the work while you make decisions. The video addresses doubts about technical difficulty, needing new tools, and safety.
1) Download Open Mousebot. 2) Create Jarvis with instructions and voice. 3) Add two crew members (researcher, writer) with single-paragraph instructions. 4) Make a group chat with Jarvis as lead and give one real job.
Open Mousebot enables users to build a free, local AI crew that can automate real work across inbox management, research, planning, and content creation. The key to trust is the 'drafts only' and 'ask me first' settings, ensuring humans remain in control.
Instructions are the key
Emphasizes that writing clear instructions is the most critical step, turning a chatbot into a trusted assistant.
03:18Crew self-checking
Relay's proactive flagging of the export bug shows how a multi-agent system can cross-verify information.
10:05Jarvis catches a false line
Demonstrates the value of a 'boss' bot that reviews work for accuracy and context.
13:09The 'before it acts' setting
This setting is crucial for trust, allowing users to control bot autonomy and ensure safety.
17:12Old way vs new way
Clearly articulates the paradigm shift from human-led work to AI-led work with human oversight.
18:12[00:02] crew sorted your inbox, [music] found the bug that kept coming up, and drafted every reply. Shall I walk you through it? That's Jarvis. I built him in one afternoon inside Open Mousebot, a brand new free open-source alternative to
[00:18] Grockbot. [music] And Jarvis doesn't work alone. He runs a whole crew of AI bots for you. One reads websites in its own browser. One writes in your voice. One turns every result into a stunning 3D scene. One clears
[00:33] your inbox. And Jarvis is the boss who hands out the work, checks it, and reports back to you out loud. Today, I'm going to show you exactly how to build him step by step on the AI tools you already use. Then, I'll put the whole
[00:48] crew to work on four real jobs from four different kinds of business, so you can see it running with your own eyes. And stick with me to the end because I'll show you the one setting that decides whether your crew works for you or just
[01:01] chats at you. By the end of this video, you'll have your own Jarvis and a team of bots that can take real work off your plate this week. Let's get into it. >> Before we begin, a quick introduction. I'm the chief of staff. The others do
[01:16] the heavy lifting. I make sure it's done properly. >> So, what is Open Mousebot? It's a chat app like the messaging app on your phone, except every contact in the sidebar is an AI bot. And every bot is a
[01:31] real AI agent with its own job, its own personality, and its own brain. That's the idea Grockbot made popular. The difference is that this one is open- source, it's free, and it runs on your own computer. Here's why that matters to
[01:45] you. Inside the AI profit boardroom, one of the questions I see over and over is some version of this. How do I stop my bots hitting their usage limits? Open mousebot plugs into the AI tools you're already logged into. If you use Clawude,
[02:01] your bots can run on your Claude. If you use Codeex or Grock, they can run on those. And it goes further than that. In the model picker on my Mac, I've got Grock, Deepseek, Muse, Kimmy, Open Code, Hermes, all sitting in one list, so each
[02:16] bot can get the brain that suits its job, and you're never stuck on one tool. It runs on Mac, Windows, and Ubuntu. Over 23,000 people have started on GitHub, and new versions land almost every day. The one I installed today is
[02:31] version 0.1.161. >> Enough about the building. Shall we build me? Step one is the install. You go to the open mousebot page on GitHub. Click the big download button for your computer and open it like any other app.
[02:46] On a Mac, you drag it into applications. The Mac version is signed and checked by Apple, so it opens normally. When it opens, you land in a clean chat window with one bot already waiting called Mousebot. Mousebot is your built-in
[03:01] We're going to build our own team next to it. Step two is creating Jarvis. You click the plus at the top of the sidebar, pick new bot, and a fresh bot appears. I rightclicked its name, hit rename, and typed Jarvis. Step three is
[03:18] the most important step, and it's the one most people skip. You open the bot settings, and you write its instructions. This is the bot's job description, and it's the part that turns a chatbot into an assistant you
[03:31] can trust. Here's what I wrote for Jarvis. You are Jarvis, my calm chief of staff with a light British manner. Short, warm sentences. Open with right Short, warm sentences. Open with right away or done when it fits. Never do the
[03:45] whole job yourself. Brief scout, quill, pixel, and relay in parallel. Check their work and hand back one tidy summary with the files. Drafts only. Never send or post. That last line is the safety net. Jarvis can draft every
[04:00] email in your inbox and nothing leaves until you say so. Then underneath there's a box called job. I typed chief of staff runs the crew. That little line shows up next to his name everywhere in the app so you always know who does
[04:14] >> Now give me a voice, would you? I'd rather not be read silently. >> Step four is the voice. This is the part people don't expect from a free app. Every bot in Open mousebot can talk. In
[04:28] the bot's voice settings, you pick a voice provider. You can use 11 Labs, Fish Audio, Grock, Gemini, OpenAI, or the voices already built into your Mac. I went with a built-in British voice called Daniel because that's free and
[04:42] it's already on every Mac. Then you switch on read replies aloud. From that moment, Jarvis speaks his answers as they arrive. You can even ring him. you can talk to your crew hands-free while you make a coffee. I'll need
[04:58] staff. One man can only do so much, even this one. >> Step five is the crew. Open mousebot has a templates page full of ready-made bots. A researcher, a writer, an inbox helper, a social media manager, an
[05:13] operations assistant, and more. You click add, and a setup window opens where you can see everything before the bot is created. its name, its mascot, its instructions, its skills, its memory, and its voice. I added four:
[05:28] Scout, the researcher, Quill, the writer, Pixel, the designer, and Relay, who runs the inbox and the follow-ups. For each one, I wiped the template instructions and pasted in one short paragraph saying exactly what that bot
[05:43] does. Scout opens real websites in its own browser and comes back with a short brief and sources. Quill writes in plain words, short lines, and saves every draft as a file. Pixel turns every result into a full screen 3D scene.
[05:59] Relay sorts messages into urgent, today, and later, and drafts a reply for each and later, and drafts a reply for each one. Step six is the brains. Click the model name at the top of any chat, and you get the picker. Jarvis runs on Grock
[06:12] you get the picker. Jarvis runs on Grock 4.7. Scout and Pixel run on DeepSeek V4 Pro. Quill runs on Muse Wonder. Relay runs on DeepSeek V4 Flash because sorting an inbox should be quick. So you get a whole team where each member has
[06:27] the right brain for its job. And when one tool gets stuck or runs out of usage, you just switch that one bot to another. That happened to me while making this video. Halfway through a job, Jarvis on Grock got stuck reading a
[06:40] file. So I opened his model picker, picked Deepseek, and he carried on. Two clicks, same Jarvis, same instructions, new brain. Step seven brings it all together. You click the plus again, pick new group chat, call it Jarvis HQ, and
[06:56] tick all five bots. Then you set Jarvis as the lead. That means when you write in the group, Jarvis answers first and decides who does what. And there's a box for group instructions that every bot reads on every turn. Mine says Jarvis
[07:12] leads. He splits every job across the crew in parallel. Drafts only. Save every file in this folder and Pixel finishes every job with a 3D scene. That's the whole build. Seven steps, all clicks and short paragraphs of plain
[07:27] English. Now, if you want this exact setup without building it from scratch, I've put it inside the AI Profit Boardroom. You get the Jarvis HQ crew for open mousebot ready to load. Jarvis's full instructions, plus scout,
[07:41] quill, pixel, and relay, each with the exact paragraph I gave them and the group brief that makes them work as one team. You also get the exact prompts I use to give the crew jobs, the same ones you're about to see, so you can paste
[07:54] them in on day one. And you get the agent OS, the system I use to run my own agents with a 30-day road map that walks you through setting up open mousebot, your bots, their voices, and their models. There are live coaching calls
[08:08] every week where you can bring your exact setup and get unstuck with people who are running bots like these right now. And there are daily tutorials on new agent tools like this one, usually the day they come out. Over 3,900
[08:22] business owners are inside and someone is always online. The link is in the comments and description or go to aiprofitboardroom.com. >> Right, then let's see what the crew can do. Job one is a software founders
[08:35] support inbox. Picture a small software company with a project board app. They shipped an update yesterday and overnight eight support emails came in. A login problem, two people saying the same export button is stuck, a question
[08:50] about adding new team seats, a feature request, a customer thinking about leaving. That's the morning every founder dreads because the first hour of the day disappears into the inbox. So, I typed one message into Jarvis HQ.
[09:05] Morning, Jarvis. The overnight support inbox is in supportinbox.md. Sort it into urgent, today, and later. Spot the bug that keeps coming up. Draft every reply. Write a short change log note for the fix and have Pixel build a
[09:20] 3D scene of the night. Go. And here's what happens next. Jarvis answers first. Right away, I'm splitting the overnight inbox across the crew and pulling the file so each brief is concrete. >> Then he writes a brief for each bot and
[09:37] lines them up. Relay builds the follow-up queue. Quill drafts all eight replies and a change log note. Scout sorts everything and hunts for the bug that keeps coming back, and Pixel builds the scene. You can watch every bot's
[09:51] status in the sidebar while it works, and you can read every brief Jarvis writes. And here's a detail worth pausing on. Relay noticed something nobody asked it to check. It flagged that the export bug was only what two
[10:05] customers reported and told Scout to confirm it before anyone calls it a known issue. That's a crew checking each other's work. About 18 minutes later, Jarvis comes back. >> Done. The overnight inbox is sorted.
[10:18] Eight replies are drafted and the night scene is in the chat above. Nothing has >> You don't get eight separate answers. You get one calm report with the files
[10:30] attached. Urgent is the Safari login loop, the two export reports, and the customer thinking of leaving because the mobile app feels slow. Today is an invoice request and an API question. Later is two feature requests. The bug
[10:45] that keeps coming up is the export button, which hangs on boards with more than 200 cards since the update. And look how careful the drafts are. They say two customers reported it because nobody has reproduced it yet. Every
[10:59] reply is saved as a draft, ready for the founder to read and send. And then pixels scene plays, and the whole night is acted out in 3D with the same red fault flaring for both customers. That's an hour of inbox work turned into a
[11:14] 5-minute read over coffee. >> Next, a new client for a marketing freelancer. Scout, you're up. >> Job two is lead research for someone who sells SEO and ads. A lot of you in the boardroom told me your biggest headache
[11:29] is getting leads and turning them into clients. And the step that wastes the most time is the homework you do before the first call. So, a new lead comes in asking for an SEO audit. To keep it real, I gave Scout a real website to
[11:44] look at, the open mousebot site itself. I tell Jarvis, "New lead for my SEO freelance business. They want an SEO audit. Have Scout open their site in its own browser and find three things I could fix." Then, Quill writes a
[11:58] one-page call brief, and Relay drafts a friendly follow-up email for after the call. Drafts only. Scout opens the website in its own browser. This is the part that makes Open Mousebot different from a normal chatbot. Every bot has its
[12:13] own computer to work on. It can be a cloud desktop, a private virtual machine, or your own Mac if you allow it. So, Scout really is reading the pages the same way you would. And the first thing it noticed is something I
[12:26] hadn't told it. The address I gave it sends visitors on to a different address. So, Scout audited the page people actually land on. About 8 minutes fixes. >> Do this first. The page uses three names
[12:41] for one product. Pick one name and use it everywhere. >> Fix two. The main heading doesn't say what the product is or that it's a team of AI bots. Fix three. The page title leads with somebody else's product name
[12:56] before it gets to its own. Quill turned that into a one-page call brief that ends with a question to ask the client. Relay drafted the follow-up email and then Jarvis did the thing a good chief of staff does
[13:09] >> before that email ever goes out. Cut the line about enjoying hearing what they're building. You haven't said that. >> Relay had written a nice line about a conversation that hadn't happened yet and Jarvis caught it. That's why you
[13:23] want a boss on the crew and not just workers. So you walk into the call already knowing the three fixes with a brief in front of you and a follow-up ready to go. That's the difference between turning up to a sales call cold
[13:36] and turning up as the expert. >> Now, a Monday, everyone's least favorite morning. Let's fix that. >> Job three is for founders and business owners. The thing I hear most inside the boardroom isn't that people lack ideas.
[13:52] It's that they're overwhelmed. Too many threads, too many tabs, too many halffinish jobs, and no clear first move. So, I gave Jarvis a founders notes for the week, 12 unread email threads, a partner demo on Thursday with no slides,
[14:09] an investor update halfwritten, a designer they might hire, a welcome email pointing at an old help page, a customer saying the mobile app feels slow, and a wish for one afternoon with no meetings. And I typed, "Jarvis, it's
[14:23] Monday and I'm overwhelmed. Read my notes. Pick my top three priorities in order. Tell me what to drop or delegate and keep it short enough to read out loud. Quill writes the plan. Relay writes the handoff notes. Jarvis checks
[14:37] both against the notes before he says a word. And then he reads you your week. word. And then he reads you your week. >> Your three priorities in order. One, the three customer threads about the export bug. Start there. Two, the partner demo
[14:51] on Thursday. The slides aren't started, so it needs time early. Three, the so it needs time early. Three, the investor update due Friday. The draft is half done. Drop this week. Hiring a designer. It isn't decided, so it can
[15:07] wait. Hand off the welcome email and a one-page note on recurring tasks. So, you only have to decide and keep one afternoon with no meetings. That's the
[15:19] moment the whole video has been building to. You give it the mess and it hands you back a plan you can hear in 30 seconds. Relay even marked its handoff notes as drafts, not sent, and pointed out which words were its own and which
[15:33] came from the notes. And Pixel built the week as a 3D scene with the three priorities lighting up in order. Content next. One idea, a whole week. Quill
[15:45] pixel with me. >> Job four is for creators. Most people who make content don't run out of ideas. They run out of time to turn one good idea into enough posts. So, I gave Jarvis one idea. Most people use AI like
[16:01] a search box, but the people getting ahead use it like a team. And I asked for a full week from that one idea. A LinkedIn post for Monday, a five slide carousel for Tuesday, a short video script for Wednesday, a thread for
[16:15] Thursday, and a newsletter intro for Friday. with one rule. Make it read like one story across the week. About seven minutes later, it was done. minutes later, it was done. >> Done. One idea, five pieces, one story
[16:28] that deepens as it goes. Each piece ends by pointing to the next. Drafts only, >> And that's what makes it different from asking a chatbot for five posts. Monday
[16:41] plants the problem. One question, one answer, stop, start again. Tuesday's carousel shows what handing over a whole job looks like. Wednesday's script puts the two ways of working side by side. Thursday names the roles, and Friday
[16:56] lands it. Every post ends with a little hook for the next one, so your audience keeps coming back all week. Pixel built the week as a 3D scene, too. Five beats lighting up one by one. Everything lands in the folder as drafts. Nothing gets
[17:12] posted. You read it, you tweak it, and your week of content is done before lunch on Monday. Now, remember the one setting I said decides whether your crew works for you or just chats at you. It's the one called before it acts. Every bot
[17:27] has it. You can choose ask me first or just do it. When a bot wants to run something on its computer, edit a file, or take an action, open mousebot shows you a little card with allow and deny. So, you decide how much freedom each bot
[17:43] gets. For drafts in a folder, I let them get on with it. For anything that leaves your business, like an email or a post, the answer is always drafts first, and you send it yourself. That one setting is why you can trust a crew like this
[17:57] with real work. >> Shall I compare our way of working with flattering. >> Here's the old way. You open one AI chat. You explain your business again. You copy the answer into another tool.
[18:12] You open a second chat for the next job. You check everything yourself, one window at a time. And when you hit a usage limit, the work stops until tomorrow. Here's the new way. You write one message to Jarvis. He splits the job
[18:27] across a crew that already knows its roles. Each bot runs on the right brain for its job on its own computer. Everything lands in one folder as drafts and you get one calm report spoken out loud if you like. The old way is you
[18:42] doing the work with AI helping. The new way is the crew doing the work while you make the decisions. Now, you might be sitting there with a few doubts. The first one is this sounds technical. Can I really set this up? Everything I
[18:56] showed you today was clicks and short paragraphs of plain English. Download, open, add bots from templates. Write what each one does. Make a group chat. If you can set up a group chat on your phone, you can set this up. And if you
[19:11] get stuck, the boardroom calls are made for exactly that moment. The second doubt is, do I need to sign up for a load of new AI tools? No. Open mousebot works with the AI tools you're already logged into on your computer, Claude,
[19:26] Codeex, and Grock, plus a long list of other engines. So, you start with what you've got and you only add more if you want to. The third doubt is, is it safe to let bots loose on my business? That's what the instructions and the ask me
[19:41] first setting are for. Your bots draft and you send. Your bots ask and you approve. And because it runs on your own computer, your chats and files live with you. So, here's what you can do this week. Step one, download Open Mousbot
[19:57] and open it. Step two, create Jarvis and paste in his instructions and give him a voice. Step three, add two crew members from the templates, a researcher and a writer, and give each one a single paragraph saying what it does. Step
[20:12] four, make a group chat with Jarvis as the lead, and give him one real job from your own inbox or your own week, one job at a time. >> And if you'd like the full crew without the fuss, I hear there's a rather good
[20:25] place for that. If you want help doing that, join the AI profit boardroom. Inside you get the complete Jarvis HQ setup for open mousebot. Jarvis plus scout, quill, pixel, and relay with every instruction, the group brief, and
[20:40] the exact job prompts from the inbox, the lead research, the Monday briefing, and the content week ready to paste. You get the agent OS and the 30-day road map that takes you from download to a crew that clears real work every morning. You
[20:54] get weekly coaching calls where you can share your screen and get your bots working with people who've already done it. And you get daily tutorials on the newest agent tools. So when OpenMousebot ships its next big upgrade, you'll know
[21:07] how to use it straight away. The link is in the comments and description or go to a profitboardroom.com. So that's how to build your own Jarvis on openmot. It's free and open source and it runs on your own computer. It
[21:22] works with the AI tools you already use, and each bot can have its own brain. Every bot gets its own computer, and Jarvis can talk. Jarvis leads a crew that sorts inboxes, researches leads, plans your Monday, and writes a week of
[21:37] content. And you stay in charge because your bots draft and ask before they act. Drop a comment and tell me, would you give your crew a job from your inbox first or a job from your content? >> I'll be here when you need me. I'll see
[21:51] >> I'll be here when you need me. I'll see you in the next one.
The video opens with a rapid-fire demonstration of Jarvis sorting an inbox, immediately showing the end result and hooking viewers with the promise of a powerful, free tool. This creates curiosity and a sense of urgency to learn how to replicate it.
I built [SOLUTION] in [TIME] using [TOOL]. Watch it [ACTION] in seconds. Here's how you can do it too.
AI-generated clip ideas for Shorts based on the transcript
The promise of building a personal AI assistant like Jarvis for free is highly appealing and taps into the fascination with AI and productivity.
▶ Play ClipControversial comparison to a popular paid tool with a free alternative sparks curiosity and debate, driving engagement.
▶ Play ClipA clear, actionable tip that promises control over AI autonomy is highly educational and shareable for anyone using AI tools.
▶ Play ClipDemonstrates a tangible, impressive real-world outcome of AI collaboration, showcasing efficiency and intelligence.
▶ Play ClipOffers a concrete, reusable prompt for content creators, providing high practical value and encouraging saves and shares.
▶ Play ClipSnapshot 2026-11-17
Performance history is still being collected — a chart appears once TubeSum has 3 or more observations for this video (1 recorded so far).
Exact daily observations; gaps mean no observation was taken that day.
| Date | Views | Avg views / hour | Channel median |
|---|---|---|---|
| 2026-11-17 | 2,579 | 515.8 | — |
Audience reaction — separate from the performance metrics above.