AI Compare Ninja
Compare AI tools. Use prompts and skills.
With local AI, the language model runs on your laptop or PC instead of in a provider's cloud. Once downloaded, much of it also works offline. Here you'll find out when that makes sense and how to get started.
You install an app such as Ollama, LM Studio or Jan and download an open model, for example Llama, Gemma or Qwen. The app then generates the answers using your processor or graphics card. Your prompts do not have to be sent to a server.
Local AI is useful when you work with confidential texts and do not want to upload them, often work without a stable internet connection, want to try AI without a subscription or account, or want to decide yourself which model and version you use.
The bigger the model, the more RAM or graphics memory it needs. A few guide values according to the providers: for Macs, Jan lists 8 GB of RAM for models with 3 billion parameters (3B), 16 GB for 7B and 32 GB for 13B. According to OpenAI, gpt-oss-20b runs with 16 GB of memory. According to Google, Gemma also comes in small sizes for laptops and phones. According to the providers, llama.cpp and whisper.cpp also run on the processor alone; a graphics card or Apple Silicon makes them faster. According to the provider, ComfyUI for images and videos runs with as little as 4 GB of graphics memory and 8 GB of RAM.
The biggest benefit: when the app and the model run offline, your prompts stay on your device. But there are limits: many apps can also connect online models. Your data then goes to that provider. According to the providers, some apps send anonymous usage data. Check the settings and turn it off if you do not want it. Your computer itself has to be well protected, for example with updates and disk encryption. Every open model has its own licence terms. Read them before using the model at work. Local models can be wrong too. Check important answers.
1. Install an app such as LM Studio, Jan or Ollama from the provider's website. 2. Download a small model that fits your memory. The apps usually suggest suitable models. 3. Ask your first question, starting with texts that are not confidential. Then check the usage data settings.
All entries in the Local AI category. Details according to the providers.
Not necessarily. According to the providers, llama.cpp and whisper.cpp also run on the processor alone. A graphics card or Apple Silicon makes them much faster.
According to the providers, the apps and models in this list are free. You only pay for your hardware and electricity.
Yes, once you have downloaded the app and the model. You need internet again for updates and new models.
That depends on the model and your computer. Small models run on normal laptops but are often weaker at difficult tasks. Test with your own tasks.
That depends on each model's licence. Some licences are very open, others have conditions. We show the licence on the model's page; this is not legal advice.