If you are new to the world of local AI, opening the Ollama model library can feel a bit overwhelming. When you look at the models, you will see different tags: Cloud, Embedding, Vision, Tools, and Thinking.
What do these words actually mean? And more importantly, which one do you need for your project?
1. Cloud Models
Most models on Ollama download directly to your computer and run using your own hardware. Cloud models are the exception.
- What they are: These are models that live on powerful external servers (the cloud) instead of your machine.
- How they work: Ollama acts as a bridge. You type your question locally, Ollama sends it to the internet, and the cloud server sends the answer back.
- Best for: People who have older computers or laptops without a strong graphics card (GPU), but still want to use huge, powerful AI models.
2. Embedding Models
These models are unique because you cannot have a normal conversation with them. If you ask an embedding model “How are you today?”, it will not answer with words.
- What they are: They turn words, sentences, or whole documents into long lists of numbers called vectors. These numbers represent the meaning of the text.
- How they work: If two sentences have a similar meaning (like “The dog chased the ball” and “A puppy ran after a toy”), the model gives them similar numbers.
- Best for: Building a local search engine or a system that reads through your private documents to find the right information (often called RAG, or Retrieval-Augmented Generation).
3. Vision Models
Standard AI models can only read and write text. Vision models give the AI a pair of eyes.
- What they are: These are multimodal models, which means they can understand both text and images.
- How they work: You can upload a photograph, a chart, or a screenshot along with your text question. The AI analyzes the pixels of the image to answer you.
- Best for: Extracting text from an old PDF, describing what is happening in a photo, or asking the AI to write website code based on a drawing you made.
4. Tools Models
If you want your AI to actually do things rather than just talk, you need a model tagged with Tools (also known as Function Calling).
- What they are: These models are specially trained to connect with external software, databases, or APIs.
- How they work: Instead of answering a question like “What is the weather in London?” with general text, the model is smart enough to say: “I need to use the Weather Tool for London.” It then outputs a clean piece of data that your computer code can easily read.
- Best for: Creating AI assistants or “agents” that can control your smart home, look up data in a spreadsheet, or send emails automatically.
5. Thinking Models
These are the newest types of models on the market. They are designed for difficult problems.
- What they are: Traditional AI models guess the next word immediately. Thinking models are trained to pause, plan, and “talk to themselves” before they give you the final answer.
- How they work: When you ask a question, you will see the model processing its thoughts in the background. It tests different ideas and fixes its own mistakes before it shows you the final result.
- Best for: Advanced computer coding, hard math problems, logic puzzles, or complex science questions. Keep in mind that they take a bit longer to reply because they spend time “thinking”!
Summary Table: Which should you choose?
| If you want to… | Choose this type: |
| Chat, write emails, or summarize text on a standard PC | Standard Text Models |
| Analyze photos, charts, or diagrams | Vision |
| Connect your AI to the internet or local apps | Tools |
| Solve complex coding bugs or logic puzzles | Thinking |
| Search through thousands of personal documents | Embedding |
Tip for Beginners: If you are just starting out and have a decent computer, look for a small, standard text model first (like Llama 3 or Mistral). Once you feel comfortable, start experimenting with Vision or Thinking models to see what your local AI can truly do!
Useful Links:
Do I need an internet connection to use Ollama models?
For most models, no. Once you download an Embedding, Vision, Tools, or Thinking model to your computer, it runs completely offline. You only need the internet to download them the first time. However, Cloud models do require a constant internet connection because they talk to external servers.
Can one Ollama model belong to more than one category?
Yes, absolutely! Many modern AI models are multi-talented. For example, you can find a model that is a Vision model and a Tools model at the same time. This means it can look at a picture and also connect to your external apps.
Are these Ollama models free to use?
Yes. Ollama itself is free and open-source, and almost all the Ollama models you can download through Ollama are free for personal use and learning. Always check the specific license of a model if you plan to use it to make money for a business.
Discover more from AI with Katarina
Subscribe to get the latest posts sent to your email.
0 Comments