4 Examples of Local AI Use Cases

For today's newsletter, I wanted to share some of the practical examples I showed during the session.

Because there's quite a bit more you can do with Local AI than just running your own version of ChatGPT.

Let's dive in!

Ready to transform artificial intelligence from a buzzword into your personal revenue generator?

HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.

Inside you'll discover:

Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.

Get Your Guide

This is probably one of the first things that comes to mind when you think about “Local AI”.

Getting your own “Local ChatGPT”.

And it makes sense. It's relatively easy to set up, which is why it was one of the first things I showed during the webinar. But there's also a limit to how useful it is (more on that in a bit).

I'm using an open-source tool called Open WebUI, connected to an open-weight AI model (Qwen 3.8) running on my NVIDIA DGX Spark.

It works pretty much like you'd expect:

You can ask questions, analyze documents, or even have the AI write and execute code to solve more complex problems (in a sandboxed environment).

You can also connect it to your own data sources and let multiple people use the same system.

The nice part is that the AI model runs on your own hardware, under your control. You don't need to send your prompts and documents to OpenAI, Anthropic or any other AI provider.

This is definitely useful – especially if you’re dealing with sensitive data – but there’s also a limit to how far a snappy chat interface can get you. There’s a limit to Productivity AI – and buying your own AI hardware just for this might not be worth the investment.

I think the more interesting opportunities start when we move beyond the chat interface and build applications that integrate deeper into your workflows and deliver specific outcomes. That's the shift from what I call Productivity AI to Engineered AI.

And to make this a little more concrete, here are three applications I've built and demoed during the webinar:

As much as I dislike “second brains” for general-purpose applications (I don’t believe in “an AI that knows your entire company knowledge”), I do believe in domain-specific second brains that organize information around relevant business areas.

In my particular case, that’s everything I've written about profitable AI over the past years, including my book, newsletters, paid workshops and other materials.

Instead of just storing all these documents in a folder, the system organizes everything into concepts and connects related ideas, similar to a wiki:

AI keeps the whole system up to date every time I write something new. And now I can interact with that knowledge using my local AI model. For example, I can ask about the difference between AI agents and workflows, look at related concepts, or find gaps in my content for things I haven't covered yet.

Granted, not everyone has a large collection of content sitting around. But imagine using something like this for your company's project documentation, legal cases, research material, or any other proprietary information.

Especially if you don't want to throw all of that data into the abyss.

The next example is probably even more practical.

This one is inspired by client work involving multiple shared email inboxes like support@ or invoices@. But I've adapted it to my own situation, where my personal inbox literally consists of five different accounts linked to different businesses and partnerships I’m involved in.

If you're managing multiple email inboxes, you know how painful it can be to keep track of everything.

Of course, Gmail, Outlook and co. have built-in AI features. But typically, every AI lives inside its own application. If you want AI to look through all your inboxes, you need something else.

I call this my AI Mailroom.

The application downloads emails from my different accounts onto my local hardware and then uses AI to process them.

It does a couple of things for me:

Plus a little bonus feature that allows me to quickly draft professional replies using very short instructions:

None of these features is particularly revolutionary, and you can find plenty of similar tools out there on the market. The benefit for me is that I can customize the system exactly to my needs and process emails locally, without sending their contents to another AI provider.

And I don't have to hand my entire email communication to one single provider.

The last example is a little bit more boring.

But potentially also the most profitable.

I was working with a customer that had a team of 15+ accountants spending days on complex account reconciliation tasks every month. Now, I'm obviously not saying that AI can replace those accountants. But there's a lot of manual work involved in matching transactions, checking balances, and figuring out which entries in which ledgers belong together.

So I built a local AI application that helps with exactly that.

The workflow is super simple:

The system can detect straightforward one-to-one matches, but also more complex n:1, or even n:n cases, where multiple transactions can match multiple other transactions across different accounts.

These results can be exported to a spreadsheet or integrated into another business system.

Of course, putting a system like this into production requires proper setup, testing and ongoing evaluation (which is the real work), but you can imagine how this might change the way your accounting team approaches monthly reconciliation.

This is what I mean by Engineered AI.

You don't need someone sitting in front of a chatbot, asking it questions and copying answers back and forth. You have an application that takes some input, performs a specific task, and delivers an outcome right back to you.

The key idea here is that you can delegate a substantial part of a repeatable business process to AI, rather than just using AI to assist someone doing that process manually.

Which is the real deal.

These are just four tiny examples of what you can do with Local AI.

In reality, there are countless things you can build that are potentially valuable for your business.

And you don't necessarily need frontier intelligence for all of them. Many individual tasks inside business processes are pretty dumb. And today’s open-weight models are already smart enough to handle a good bunch of them.

Of course, turning these applications into reliable production systems is a different story.

But I think that's where Local AI gets interesting. Not just running your own ChatGPT, but building applications that actually do something useful.

You “just” have to do it.

See you next Saturday,
Tobias