--- title: "Local AI Models · AssistantGeneral" description: "Runs the AI on your own computer, offline: a built-in model, any GGUF file you pick, a check that it fits your hardware, and faster writing on the default." url: "https://royalsoftworks.com/products/assistant-general/features/local-ai-models/" source: "https://royalsoftworks.com" format: "markdown" note: "Markdown rendering of the HTML page at `url`. Same content, same canonical URL." --- Features · The AI itself # Run the AI on your own computer. Runs the AI on your own computer, offline: a built-in model, any GGUF file you pick, a check that it fits your hardware, and faster writing on the default. [Editions](https://royalsoftworks.com/products/assistant-general/editions/): Every edition, including Free. Free cannot change the system prompt, the temperature or network sharing. ## Local AI models, feature by feature Each comes with a small drawing of how it works. They are diagrams, not screenshots; numbers in them are examples. - ### A built-in model, one download The first start offers a ready model, its vision part and a small drafting helper, all fetched from Hugging Face. Nothing else has to be installed. For your IT team: Google Gemma 4 E4B (quantised GGUF), its image projector and a 60 MB multi-token drafter. Diagram: First start → Hugging Face → Model, vision + drafter. - ### Any model you choose Point it at any GGUF model file, or search Hugging Face from inside the app: Llama, Mistral, Qwen, Phi, Gemma, DeepSeek and hundreds more. Diagram: A GGUF file and Hugging Face search come together as Your model. - ### Will it fit? Before you download The model picker says whether a model fits in your graphics memory, needs partial offload, runs on the processor only, or is too big for this machine. Diagram: Example: how much of each model fits in graphics memory — fits, partial offload, or too big for this machine. - ### Matched to your graphics card The engine build is chosen for the hardware it finds, and pinned to a version we tested. It can also be installed from a file you bring in. For your IT team: llama.cpp builds for NVIDIA (CUDA 12.4 / 13.4), other graphics cards (Vulkan) and CPU-only. Diagram: Your hardware leads to NVIDIA → CUDA, Other GPU → Vulkan, No GPU → processor. - ### A speed check on first start A short benchmark measures this computer once, and the app sizes its work to the result, so a laptop is not asked for what a workstation can do. Diagram: Short benchmark → Tokens per second → Work sized to match. - ### Faster writing, same answer On the built-in model a small helper drafts the next words and the main model checks every one, so the text is unchanged. On one test machine: about 1.6× faster on an NVIDIA card, 1.3× on Vulkan, 1.1× on the processor. Diagram: Writing speed-up on the built-in model, measured on one test machine: about 1.6 times on an NVIDIA card, 1.3 on Vulkan, 1.1 on the processor. - ### Reads pictures Vision models take photos and scanned pages directly. Switch the vision part off to save memory when you only need text. Diagram: Photo or scan → Vision part → The model reads it. - ### A second engine KoboldCpp can stand in for llama.cpp, switched in Options. Both are downloaded for you and kept to a tested version. Diagram: Your model leads to llama.cpp, KoboldCpp. - ### Your settings, if you want them Context size, the engine's own command-line options (with the translated command shown), a temperature that adapts to each request or stays fixed, and how long the model thinks before it answers. Diagram: Context size, Engine options, Temperature, Thinking time. - ### Share it on your network Optionally let other computers on your local network use the model this machine runs. Diagram: This PC's model leads to Desk 2, Desk 3, Desk 4. ## Goes well with - [Cloud AI providers](https://royalsoftworks.com/products/assistant-general/features/cloud-ai-providers/) - [Model cluster](https://royalsoftworks.com/products/assistant-general/features/model-cluster/) - [Offline mode](https://royalsoftworks.com/products/assistant-general/features/offline-mode/) [Next Cloud AI providers](https://royalsoftworks.com/products/assistant-general/features/cloud-ai-providers/) [All 31 feature classes](https://royalsoftworks.com/products/assistant-general/features/)