← HomeGUIDE 01 / GETTING STARTED

Install Local Idea Studio and run your first GGUF model

A step-by-step Windows installation guide for Local Idea Studio, including SHA-256 verification, model storage, Hugging Face downloads, and GPU setup.

Before installing

Local Idea Studio v0.1.5 supports Windows 10 and 11 on x64 computers. The application does not bundle an AI model, so leave enough free storage for the GGUF file you intend to download. An 8 GB system can run small quantized models; larger models and longer contexts require more RAM or VRAM.

The current installer is not digitally signed. Download it only from the official GitHub release and verify its checksum before running it.

1. Download and verify

  1. Download Local Idea Studio v0.1.5 for Windows.
  2. Open PowerShell in the download folder.
  3. Run the command below and compare the result with the release checksum.
Get-FileHash -Algorithm SHA256 ".\Local.Idea.Studio-0.1.5-win-x64.exe"

Compare the result with the SHA-256 shown on the v0.1.5 GitHub release. If even one character differs, do not run the file.

2. Complete setup

  1. Open the verified installer.
  2. Read the license agreement and select acceptance only if you agree.
  3. Choose an installation folder or keep the suggested per-user location.
  4. Finish installation and launch Local Idea Studio.

Windows SmartScreen may warn about an unknown publisher because the early-access installer is unsigned. A warning is not proof of malware, but it is why checksum verification matters.

3. Choose model storage

New installations store downloaded models in Documents\Local Idea Studio\models. Existing users retain their selected model folder after the rename. You do not need a D: drive. Open Settings → Storage to select any available local folder, then Local Idea Studio immediately uses that folder for new downloads and model scanning.

4. Download a GGUF model

  1. Open Models.
  2. Search for a model family such as Qwen, Llama, Gemma, Mistral, Phi, or TinyLlama.
  3. Open a GGUF repository and inspect the exact file size.
  4. For a first test, choose a smaller Q4_K_M or Q5_K_M file that comfortably fits your memory.
  5. Watch progress in Models or Downloads, then load the completed file.

5. Enable acceleration

In Settings → Model loading, keep GPU backend on Auto-detect and GPU layers on all. Auto prefers CUDA on a supported NVIDIA RTX/GTX GPU, then Vulkan when supported, then CPU. Reload the model after changing the backend and confirm the active backend in Runtime info.

6. Start a local chat

Open Chat, enter a prompt, and watch the generated token count and tokens per second. Prompts and inference remain on the computer; internet access is used when you search or download from Hugging Face or open an external link.