Local AI runs model inference on your own device. After downloading software and models, supported local workflows can work offline; cloud features and extensions may still send data. RAM is system memory, VRAM is GPU memory. CPUs can run some models more slowly; a compatible GPU can accelerate supported workloads. Larger models require more memory. Quantization uses fewer bits to reduce size, trading some accuracy for easier running. Start small, check model licenses and measure on your actual machine.
Featured starting points
Explore our collection of useful tools and applications. This selection is regularly reviewed and expanded.
Run compatible language models locally; explicitly distinguish local models from optional cloud models.
Platforms
Windows · macOS · Linux
Guidance & sources
Start with a small local model and check whether cloud functionality is enabled.
Disable optional cloud features when you require a local-only workflow. Model licenses are separate.
Local AI runs model inference on your own device. After downloading software and models, supported local workflows can work offline; cloud features and extensions may still send data. RAM is system memory, VRAM is GPU memory. CPUs can run some models more slowly; a compatible GPU can accelerate supported workloads. Larger models require more memory. Quantization uses fewer bits to reduce size, trading some accuracy for easier running. Start small, check model licenses and measure on your actual machine.
Explore a graphical workflow for downloading and running compatible local language models.
Platforms
Windows · macOS · Linux
Guidance & sources
Check the model, runtime compatibility and your device memory before a large download.
Edition and requirements can change; check the official source before installing.
Local AI runs model inference on your own device. After downloading software and models, supported local workflows can work offline; cloud features and extensions may still send data. RAM is system memory, VRAM is GPU memory. CPUs can run some models more slowly; a compatible GPU can accelerate supported workloads. Larger models require more memory. Quantization uses fewer bits to reduce size, trading some accuracy for easier running. Start small, check model licenses and measure on your actual machine.
Build node-based image generation, image processing and supported video workflows.
Platforms
Windows · macOS · Linux
Guidance & sources
Capabilities depend on installed models, custom nodes and workflows; ComfyUI alone is not a video generator. Local core is free; optional cloud/partner nodes may cost money. Stable Diffusion model licenses differ.
Edition and requirements can change; check the official source before installing.
Local AI runs model inference on your own device. After downloading software and models, supported local workflows can work offline; cloud features and extensions may still send data. RAM is system memory, VRAM is GPU memory. CPUs can run some models more slowly; a compatible GPU can accelerate supported workloads. Larger models require more memory. Quantization uses fewer bits to reduce size, trading some accuracy for easier running. Start small, check model licenses and measure on your actual machine.
Chat with supported local language models or optional remote providers.
Platforms
Windows · macOS · Linux
Guidance & sources
Download a compatible model and check RAM/GPU and its license. Local mode can work offline after downloads; enabling remote providers changes data processing and may cost money.
Edition and requirements can change; check the official source before installing.
Local AI runs model inference on your own device. After downloading software and models, supported local workflows can work offline; cloud features and extensions may still send data. RAM is system memory, VRAM is GPU memory. CPUs can run some models more slowly; a compatible GPU can accelerate supported workloads. Larger models require more memory. Quantization uses fewer bits to reduce size, trading some accuracy for easier running. Start small, check model licenses and measure on your actual machine.
Run supported language models locally and explore local document chat.
Platforms
Windows · macOS · Linux
Guidance & sources
Download a compatible model and check RAM/GPU and its license. Local mode can work offline after downloads; enabling remote providers changes data processing and may cost money.
Edition and requirements can change; check the official source before installing.
Local AI runs model inference on your own device. After downloading software and models, supported local workflows can work offline; cloud features and extensions may still send data. RAM is system memory, VRAM is GPU memory. CPUs can run some models more slowly; a compatible GPU can accelerate supported workloads. Larger models require more memory. Quantization uses fewer bits to reduce size, trading some accuracy for easier running. Start small, check model licenses and measure on your actual machine.