Skip to main content

Working with local AI models: easier than you think

You want to ask AI questions about confidential internal documents. Or use AI to draft a contract. Or prepare a board report on sensitive topics. You hesitate: Can you use ChatGPT for this? You don’t want this information to be used to train AI models.  

There is an alternative that many people are not yet familiar with: AI models that run on your own computer. Without your data ending up somewhere. Free too. It’s a misconception that technical knowledge is required for this. You can get started within half an hour!  

9 minutes15 jun `26

What is a local AI model?  

Local AI models are programs that run entirely on your own computer, without an internet connection. These programs work the same way as online AI tools like ChatGPT, Gemini, or Claude: you have a chat window where you can ask your question or give your task. You can also upload files.  

There are local AI models for various applications: text, image, video, audio, speech, etc. In this article, we primarily describe text applications because they are the most commonly used.  

Screenshot van het eindresultaat
Screenshot of local AI model Ollama

What can you do with local AI models? 

Local models work best when you give them a concrete and not too complex task or assignment. For example:  

  • Rewriting a grant application 
  • Elaborating notes on confidential information  
  • Adjusting or drafting a contract with a supplier  
  • Asking questions about internal confidential documents (project plan, grant application, employee handbook) 
  • Elaborating notes from a performance review 

Where local AI is less strong: tasks that require a lot of creativity, complex reasoning, or current knowledge. Local AI is also less strong in fetching information because its database is smaller than online models. For extensive data analysis or strategic advice, it’s wise to use a powerful online model.  

Local AI can thus be an alternative for specific tasks. For a large portion of daily writing and thinking work, it is good enough. 

The three most important programs 

A program is the environment in which you work, comparable to a media player. An AI model is the content you load into it, comparable to a music file. 

There are three programs that are easiest to start with. Within these programs, you can easily download the most commonly used AI models.  

  1. Ollama

    Ollama is the most commonly used option. You install the program, choose an AI model via the menu, and can get started right away.  

  2. Jan.ai.

    Just like Ollama, a user-friendly and simple program.  

  3. LM Studio

    LM Studio has more extensive features than the other two tools. This is an advantage if you have more technical knowledge, but the interface is (therefore) less user-friendly.  

Which AI models are available? 

The most commonly used and free open-source language models at the moment (summer 2026) are: 

  1. Llama 3

    Created by Meta. Good at many tasks. 

  2. Mistral

    A French model, compact and fast.  

  3. Gemma 4

    Created by Google. Small but powerful. 

  4. Qwen

    From the Chinese company Alibaba. Strong in multilingual tasks. 

New open-source language models are regularly launched, and their quality is improving. So make sure to stay updated on new developments. Language models are offered in different sizes: the larger the model, the ‘smarter’ it is. But also, the more computing power your computer needs.  

How much RAM do you need? 

The size of the local AI models you can use depends on the RAM of your laptop or phone. The size of the language model is expressed in billion (letter B) parameters. A rule of thumb is: approximately 1 GB of RAM per billion parameters. A model with 7B parameters therefore requires roughly 7 GB of RAM.  

How to get started 

Want to start working with local AI models yourself? Follow this simple step-by-step plan: 

Step 1. Choose a program and install it. Ollama is the most commonly used.  

Step 2. Open the program and search for a model. Start with Llama 3, Gemma 4, or Mistral. Choose the largest possible model that works with your RAM.   

Step 3. Download the model, it will be stored on your computer. 

Step 4. Ask a question or give a task, just like with ChatGPT or Copilot. 

Step 5. Experiment and discover what works and what doesn’t.  

For advanced users: Hugging Face 

Want to use more specific models? Then Hugging Face is interesting. It’s a library with thousands of AI models for specific purposes. You do need some technical knowledge to work with it, but the possibilities are vast. Visit: huggingface.co

Conclusion 

Make a conscious decision: when do you use local models, and when do you choose an online service? For high quality, complex analyses, or tasks where you need the best quality, online models are better. For sensitive or confidential tasks — personnel information, grant applications, contracts — local is the logical choice.  

Document this choice in an AI policy. As an organization, make agreements about when employees use which type of AI. This provides clarity and helps everyone work responsibly. 

The threshold is lower than you think. You can get started within half an hour! 

Use AI more effectively in your work?

Do you already have some basic knowledge about AI and want to take the next step in working with generative AI tools? During the workshop 'Deepening AI for the cultural sector' by DEN and Innovation:Lab, you will not only learn what is possible with AI tools, but especially how to apply AI iteratively. 

View the workshop

Share this news article

What are the advantages and disadvantages of local AI models?

Local AI has several clear advantages, but also disadvantages. An overview: 

Advantages of local AI models 

- Privacy: your data always remains on your computer and is not used for model training. You can therefore work with sensitive information. 

- No costs  

- Offline: works even without an internet connection. 

- Control: you are not dependent on a company that can change rules or prices. 

Small local AI models generally use less energy than large online AI models. However, the environmental impact of AI depends on so many other factors that we dare not call this an advantage.  

Disadvantages of local AI models 

- Less ‘smart’: local models are generally less advanced than large online services like ChatGPT or Claude.  

- Current information: local models do not contain the most current information and cannot fetch it from the internet.  

- Hardware: for larger models and more complex tasks, you need a powerful computer. In practice, this will be the biggest bottleneck.  

- No automatic updates: you have to fetch the latest version of the language model yourself.