[00:02] >> I am running 100% offline [music] on your Mac, sir. >> That's Jarvis Local Core, a personal AI [music] assistant that runs completely offline on my own Mac. Local brain, local voice, local memory. What does my [00:17] >> [music] >> You wake at 6:00, gym at 7:00, deep work from 9:00 to 12:00, no meetings, then record videos [music] in the afternoon. Walk after lunch, sir. >> He pulled that straight from my own [00:31] notes on my own Mac. And look at the dashboard, a glowing reactor in the middle, a 3D galaxy of my notes on the right, and every AI model on my computer lined up on the left. In this video, I'll show you what it does, the exact [00:46] prompts I use with it, and how to run it yourself in about 2 minutes. So, here's what I wanted. Jarvis from the films, but private. Client notes, business plans, my daily routine, none of that going to a server somewhere. So, the [01:00] whole thing runs on Ollama, which runs AI models on your own computer. My Mac is an M4 with 16 GB of memory, a normal machine. And it's one small project, one server file, and one web page. No accounts, no API keys, nothing to [01:16] install from NPM. Let's walk through it. On the left is the neural core. Every model I've downloaded with Ollama. Gemma 3 at 4 billion parameters, Qwen 2.5 at 7 3 at 4 billion parameters, Qwen 2.5 at 7 billion, GPT-OSS at 20 billion. Each [01:31] card shows the size, whether the model can use tools, whether it can see images, and how much of your memory it fills. Blue bar, it fits comfortably. Orange bar, it takes most of your machine. I keep him on Qwen 2.5. It's [01:44] quick, it can use tools, and it leaves plenty of memory for everything else on my Mac. In the middle is the reactor. Cyan when he's waiting, amber when he's thinking, red when he's listening. And when he talks, the rings move with his [01:58] voice. That voice is Kokoro, an open-source British voice. Runs on your Mac, so it starts instantly. And he speaks each sentence as soon as it's whole reply. How's the system holding up? [02:14] >> The system is operating efficiently with adequate resources. adequate resources. >> Down here on the left is telemetry. CPU, memory, tokens per second, and how long the first word took. On my Mac, Kwin [02:27] runs at about 20 to 30 tokens a second. If you want my best setups for local AI, prompts I use with it every day, they're inside the AI Profit Boardroom. You get [02:39] the full Jarvis project, the second brain template, and the faceless video skill I used to make this video. There's a short guide for each one, so you can set it up without guessing. And there's a community of people building their own [02:51] versions. So, if you get stuck, someone has usually solved it already. The link is in the comments and the description, or go to aiprofitboardroom.com. Now, my favorite part, the second brain. This galaxy on the right is my notes. [03:07] Every dot is a markdown file. Business notes, content plans, ideas, my team, my routine. Each color is a folder. Jarvis turns every note into an embedding with a small local model called MX by embed large, then plots them in 3D. Notes [03:24] about similar things sit close together. You can drag the galaxy to spin it, and hover over a dot to see the file. Watch what happens when I ask a question. What are my Q4 goals? >> Q4 goals are to publish three YouTube [03:38] videos a day, grow the community, launch an AI assistant course, and move routine AI work onto local models. >> The Q4 goals note lit up. A line shot out to it from the center, and he answered from that note. I never told [03:53] him which file to open. What's on my video ideas list? >> Video ideas include comparing a llama to cloud APIs, showcasing 10 everyday uses for private AI assistant, and exploring unified memory requirements for local [04:07] LLMs. >> Next, memory. Remember that I record videos at 2:00 p.m. >> Noted, sir. I'll remember that you record videos at 2:00 p.m. >> That fact is saved to a file on my Mac, [04:21] and it joins the galaxy as a new red dot. So, it's still there after a dot. So, it's still there after a restart. When do I record videos? >> You record videos at 2:00 p.m., sir. >> Now, tools. On a model that supports [04:36] tools, like Qwen 2.5, Jarvis can do things as well as talk. Set a 25-minute focus timer. >> A timer for 25 minutes has been set, sir. >> A timer appears at the top and counts [04:51] down. When it ends, he tells you out loud. Write a note called client onboarding checklist with the steps for welcoming a new client. >> The client onboarding checklist has been created and saved. [05:05] >> A new markdown file and a new dot in the galaxy. He can also open apps on your Mac. Say, open calendar, and calendar opens. Every tool call shows up in the [05:17] transcript on the right in amber. The tool name, what it was given, and what came back. So, you always know what he did. And at the top, there's a privacy bar. It says offline. It counts every call to a local model. Cloud calls stay [05:32] at zero, and data sent stays at zero bytes. So, what machine do you need? The model files tell you. On a llama, Qwen 2.57B 2.57B is 4.7 GB. Gemma 34B is 3.3. GPT-OSS 20B [05:49] is 4.7 GB. Gemma 34B is 3.3. GPT-OSS 20B is about 14. On my 16 GB Mac, the four and seven billion models feel smooth. 20 billion runs, but it fills most of the memory, which is why its bar is orange. More unified memory lets you load a [06:03] bigger brain. Here's how to run it. First, install a llama from a llama.com. Then pull three models in your terminal. A llama pull Qwen 2.57b, [06:15] A llama pull Qwen 2.57b, a llama pull Gemma 34b, and a llama pull MX by embed large, which powers the second brain. Then go into the Jarvis folder and type node server.js. Open your browser at localhost:7777. [06:31] You'll see the boot sequence, then the dashboard. The first thing I'd do, open the brain folder and drop in five notes about your business, your goals, your clients, your content plan, your team, your weekly routine. Plain markdown [06:44] files. They appear in the galaxy as soon as Jarvis starts. Then ask him about your goals and watch the right note light up. Tell him three facts about how you work, starting each one with remember that. Now he knows you. And to [06:59] talk instead of type, hold the spacebar or click the mic, and let go when you're done. Thanks, Jarvis. That's all for now. >> You're welcome, sir. Is there anything else you need assistance with in the [07:12] future? >> So that's Jarvis local core, a private assistant with a second brain, memory, tools, and a voice all on your own Mac. If you want the full Jarvis project, the second brain template, and the prompts [07:25] from this video, they're inside the AI Profit Boardroom, along with the faceless video skill I used to make this. The link is in the comments and the description. Or go to AI Profit Boardroom.com. [07:37] Thanks for watching, and I'll see you in the next one.