Speaker A
This little USB drive has a full AI model on it. No internet, no account, and no filter. I just plugged it in, doubleclicked once, and it's running.
A walkthrough of running local, offline AI tools (chat, image generation, speech-to-text, text-to-speech) from a portable USB drive across Windows, Mac, and Linux, with setup instructions.
Key Takeaways
- A portable USB drive can host a complete offline AI studio that works across Windows, Mac, and Linux.
- After initial setup, the system runs without internet, with all data and models stored locally on the drive.
- The tool bundles chat, image generation, speech-to-text, and text-to-speech in one open-source application.
What the video covers
- Presents a portable USB drive setup that runs AI models fully offline, with no account, cloud, or internet needed after initial setup.
- Demonstrates local chat with selectable models, optional 'thinking' mode, and file attachments, with web search disabled to show offline operation.
- Claims the chat models answer sensitive questions without refusals, touting the lack of company-imposed filters.
- Shows local image generation with adjustable steps, sampler, and seed, with history saved to the drive.
- Demonstrates speech-to-text transcription of recorded or uploaded audio, with selectable spoken language.
- Demonstrates text-to-speech with male and female voices in American and British accents.
- Provides step-by-step setup: format a 32GB+ USB 3.0 drive as exFAT, download the open-source project from GitHub, and launch via a platform-specific script.
- Explains downloading models from an in-app catalog, including hardware-based recommendations and an 'uncensored' filter option.
- Demonstrates the same USB drive working across Windows, Mac, and Linux, with models and chat history persisting between machines.
- Notes automatic GPU usage across Nvidia, AMD, Intel, and Apple Silicon hardware.
Chapters
- 00:00Introduction: Portable Offline AI on a USB Drive
- 00:24Platform Support and Hardware Acceleration
- 01:24Chat Demo with Offline Operation
- 02:13Image Generation Demo
- 02:58Speech-to-Text Demo
- 03:24Text-to-Speech Demo
- 03:45USB Drive Setup on Windows
- 04:51Downloading Models
- 05:45Using the Same Drive on Mac
- 06:22Using the Same Drive on Linux and Wrap-up
Full Transcript — Download SRT & Markdown
Speaker A
It's all on my own hardware, so nothing goes to the cloud, and your chats stay right on this drive. And since everything runs locally on your machine, there are no company filters. Just ask whatever you want and get straight
Speaker A
answers without the usual refusals or lectures. And it's not just chat. The same app can generate images, turn your voice into text, and read answers back out loud, all offline. And you are not stuck on one operating system. It works on
Speaker A
Windows, Mac, and Linux. And it uses your graphics card automatically, whether that's Nvidia, AMD, Intel, or Apple Silicon. So, this is portable local studio. It's completely free and open source. And once it's on this drive, it's all yours. I put a lot of
Speaker A
hours into building this, so if you find it useful, please consider subscribing to the channel. It really helps me a lot. All right, now I'll show you what it can do first, and then I'll show you the installation. Let's start with chat.
Speaker A
Let me turn off my Wi-Fi first. Here we have the option to select a model. Once you select it, it loads instantly. We also have the option to attach a file, enable web search, and turn on thinking. Web search needs the
Speaker A
internet, so I'll leave that off for now. But everything else runs completely offline. Let me turn on thinking and ask it something.
Speaker A
You can see it's thinking first, working through the problem, and then it gives the answer. And remember, Wi-Fi is off.
Speaker A
This is all running from the USB drive. Now, let me attach a file and ask it a question.
Speaker A
And there you go. It reads the file and gives me an answer without anything leaving my computer.
Speaker A
And here's the no filter part. Most AI apps would refuse a question like this, but it just answers with no lecture at all.
Speaker A
Now, let's move to image generation. I'll switch over to the image workspace. And just like with chat, I can pick an image model here. Once I select one, it starts loading.
Speaker A
Once the model is loaded, I can just type in what I want to see. Let's try something crazy.
Speaker A
And there you go. The image is generated right here on my machine with no internet at all. And if you want more control, you can change the steps sampler and seed. If you're new, just leave the defaults. They work well. And
Speaker A
everything you create is saved in your history together with the prompt and settings all on the USB drive.
Speaker A
Next, speech to text. Just like before, I pick a model here and it loads up. Now I can record my voice right here or drop in an audio file like a voice note or a lecture. Just hit transcribe and it
Speaker A
transcribes your audio locally in just a few seconds. And there it is, my voice turned into text. You can copy it or save it as a file. And you can also pick the spoken language so it works for more than just
Speaker A
English. Now let's try texttospech. First, I'll select a voice model and it loads in just a few seconds. Now, I'll paste in some text and pick a voice.
Speaker A
There are male and female voices with American and British accents. Once you've picked one, just tap speak and it will turns your written text into actual speech.
Speaker A
And there it is, a natural sounding voice generated right on my machine with no internet and no voice subscription.
Speaker A
All right. Now, let me show you how to set it all up on your own USB drive.
Speaker A
I'll do it on Windows, but the process is almost the same on Mac and Linux.
Speaker A
First, you need a USB drive. I recommend at least 32 GB and USB 3.0 or faster for good speed. Before anything else, format the drive as XFAT. That way, it works on Windows, Mac, and Linux.
Speaker A
Next, head over to the GitHub page. I'll leave the link in the description. Click the green code button, then download zip and extract it straight onto your USB drive.
Speaker A
Now, let's launch it. On Windows, just doubleclick the Windows.bat file. The first time it needs an internet connection to set things up. Everything stays on the USB drive. So, let's wait for it to finish.
Speaker A
And there you go. Our local studio opened in the browser automatically. Now, let's head over to the models page to download AI models. Here, you can browse image, text, speech, and voice models. It also recommends models that best fit your hardware. Want no filter
Speaker A
answers? Just tap the uncensored filter. Once you find one you like, just hit download and it saves straight onto the USB drive.
Speaker A
Once it's downloaded, you'll see it right here in chat. Just select it and you can start chatting.
Speaker A
You can do the same for image generation. I recommend Epic Realism. It gives really realistic results and it has no built-in filter either.
Speaker A
The model is downloaded. So, let's give it a try. And there it is. Image generation is working perfectly right from the USB drive.
Speaker A
Now, let's take this same USB drive and plug it into a Mac. On Mac, open a terminal in the folder and run the Mac script. The first time on a Mac, it needs the internet for a minute to set up. The models are already
Speaker A
on the drive, so there's no need to download them again. And there it is. The studio opens in the browser with all the same models. Even our chats and images from Windows are still here. Let's quickly test text
Speaker A
generation with the Gemma 4 model we downloaded on Windows. And yes, it's running on a Mac as well.
Speaker A
Now, let's take this same USB drive and plug it into a Linux machine. On Linux, open a terminal and run the Linux script. It needs internet for the first setup, but your models are already on the drive.
Speaker A
And there it is. The studio opens in the browser with all the same models. And again, our chats and images are all still here. Let's quickly test text generation with the same Gemma 4 model.
Speaker A
And yes, it's running on Linux, too. One USB drive, three operating systems. So, that's it. A full AI Studio on a single USB drive. Chat, images, voice, all running offline on Windows, Mac, and Linux. If you found this video helpful,
Speaker A
please consider subscribing. It really helps the channel. Thanks for watching, and I'll see you in the next one.
Topics:local LLMoffline AIUSB AIportable AIself-hosted AIimage generationspeech-to-texttext-to-speechopen sourceGemma
