DocsGet started

Installation

Bud Decision Studio is a desktop app for Windows, macOS and Linux. Setup checks your computer, installs the engine that runs the models, and then lets you choose which models to download. Nothing else needs to be installed first.

What your computer needs#

Supported
Windows Windows 10 or 11, 64-bit (Intel or AMD processor)
macOS macOS 12.3 or later on Apple Silicon (M1 or newer). Intel Macs are not supported, because current PyTorch builds no longer run on them.
Linux 64-bit Intel or AMD, or ARM64 (including the NVIDIA GB10 in a DGX Spark). The app's window needs WebKitGTK 4.1, which the .deb and .rpm packages install for you.

A GPU is optional. Setup finds the one you have and installs the matching engine:

Your hardware Models run on
NVIDIA GPU (GeForce, RTX, GB10, data-centre cards) CUDA 12.6, 12.8 or 13.0, chosen from your driver
Mac with Apple Silicon the Apple GPU, through Metal
Intel Core Ultra or Arc graphics the Intel GPU (XPU)
AMD GPU on Linux ROCm. Experimental: some models may not load.
Anything else the processor. Small models answer in about a second; large ones are slow.

Disk. The engine (a private copy of Python, PyTorch and the model libraries) takes 1.5 GB for the processor, 1.6 GB on a Mac, 3 GB for an Intel GPU and about 4.5 GB for NVIDIA or AMD. Each model adds its own download, from 0.6 GB to 24 GB.

Memory. A model uses memory only while it is loaded, and you can eject it at any time. The smallest needs under 1 GB; the largest, Jev-Omni, needs about 26 GB and a GPU.

Model Parameters Memory when loaded Download Reads
Julia 1 144M 0.8 GB 0.6 GB text
Laya Multilingual 322M 1.0 GB 0.7 GB text
Laya 421M 1.2 GB 0.8 GB text
Laya Typed-Decisions 421M 1.2 GB 0.8 GB text
Kev 0.5B 0.5B 1.3 GB 1.1 GB text
GLiNER2.5 Decide 340M 2.0 GB 2.0 GB text
Kev 4B 4B 9.5 GB 9.5 GB text
Lev 4B 9.5 GB 9.6 GB text
Intern-Decision 4B 4B 10 GB 9.1 GB text, images
CLM 8B 8B 17 GB 16.5 GB text
Jev-Omni 12B 26 GB 24 GB text, images, audio, video

You do not need all of them. One small model, such as Laya, is enough to try everything in this documentation. Models explains what each one is good at.

Install with one command#

The one-line installers fetch the latest release, install it in the usual place for your system and open the app. They also avoid the warnings that macOS and Windows show for apps downloaded through a browser.

Run the command in PowerShell. It installs the app for your user only (no administrator rights are needed), adds it to the Start menu and opens it.

Terminal
irm https://raw.githubusercontent.com/BudEcosystem/Bud-Decision-Engine/main/get.ps1 | iex

Download the installer yourself#

Every release is on the releases page. For version 0.3.0:

File Size
Bud.Decision.Studio_0.3.0_x64-setup.exe 15.1 MB
Bud.Decision.Studio_0.3.0_x64_en-US.msi 22.0 MB

Run either one. The app is added to the Start menu. The installers are not signed yet, so Windows SmartScreen may warn you: choose More info, then Run anyway.

The first launch#

The first time the app opens, it walks you through four short screens. Later launches go straight to the studio.

1. It checks your computer#

Setup lists your system, its graphics and processor, how much memory each has, and the free disk space. Choose Continue.

The welcome screen on an NVIDIA GB10: setup has found the GPU, the processor and 96 GB of free disk.
Fig. 1The welcome screen on an NVIDIA GB10: setup has found the GPU, the processor and 96 GB of free disk.

2. You choose where models run#

The recommended choice is selected: the GPU when there is one. Each choice says how fast it is and how much it downloads. Choose Install.

Where should models run? Here the GPU engine is PyTorch for CUDA 13.0, about 4.4 GB; the processor-only engine is about 1.5 GB.
Fig. 2Where should models run? Here the GPU engine is PyTorch for CUDA 13.0, about 4.4 GB; the processor-only engine is about 1.5 GB.

3. It installs the engine#

Setup prepares Python, installs the PyTorch build for your choice and the model libraries, then checks that the device works. This takes a few minutes on a fast connection; keep the app open until it finishes. Details shows the full log. If a step fails, Try again starts it again. Setup never changes anything outside the app's own folder.

Installing the engine. Each step is ticked off as it finishes.
Fig. 3Installing the engine. Each step is ticked off as it finishes.

4. You choose which models to download#

When setup is done, choose Open Bud Decision Studio. The studio opens with a list of models, two of them ticked because they suit your computer. Tick the ones you want and choose Download. They download one at a time, smallest first, in the background, so you can start as soon as the first is ready. Not now skips this step; the Models page offers the same list later.

The model list. On first launch it is titled Choose models to download; later you open it with Download models on the Models page, where models already downloaded are marked On thi
Fig. 4The model list. On first launch it is titled Choose models to download; later you open it with Download models on the Models page, where models already downloaded are marked On this computer.

Where your files are#

The app keeps everything it creates in one folder per user:

Folder
Windows %APPDATA%\ai.bud.decisionstudio
macOS ~/Library/Application Support/ai.bud.decisionstudio
Linux ~/.local/share/ai.bud.decisionstudio

Inside it:

Path What it holds
engine-env/ the engine: Python, PyTorch and the model libraries
data/studio.db History, templates and their versions, test examples and settings, in one SQLite database
data/blobs/, data/uploads/ files you attached to decisions (images, audio, video)
data/config.json where models run (the choice you made in setup)
data/secret the key used to fingerprint variables marked sensitive; keep it with the database
data/backups/ a copy of the database made before each upgrade changes it; the three most recent are kept
logs/studio.log the studio server's log

The models themselves are not in this folder. They stay in the standard Hugging Face cache (~/.cache/huggingface/hub, or C:\Users\<you>\.cache\huggingface\hub on Windows), so other tools on your computer can share them. It follows Hugging Face's own settings, such as the HF_HOME environment variable, and the Models page shows the folder in use.

The studio serves its API at http://127.0.0.1:8420. If that port is taken, it uses the first free port from 8421 to 8440 and keeps it for later launches. The API page always shows the address in use.

Updating#

Install the new version the same way you installed the first one: run the one-line command again, or install the new file over the old one. Your engine, models, templates and History stay where they are, and setup does not run again.

When a new version changes the History database, the studio copies the database to data/backups/ before it upgrades it. A database written by a newer version than the one you are running opens read-only, so nothing is lost if you go back to an older version.

To change where models run (for example, after adding a GPU), open the System page and choose Run setup again.

Uninstalling#

Open Settings, then Apps, find Bud Decision Studio and choose Uninstall.

Removing the app leaves your data and models in place, so a reinstall picks up where you left off. To remove them too:

  • Delete the app's data folder (see Where your files are). This removes the engine, History, templates and settings.
  • Delete models from the Models page before uninstalling (select a model, open its Load tab and choose Delete files), or delete their folders from the Hugging Face cache. Other tools that share the cache may be using the same files.
Terminal
Linux
sudo apt remove bud-decision-studio      # Ubuntu, Debiansudo dnf remove bud-decision-studio      # Fedora# the AppImage, or wherever you kept it, and its menu entryrm ~/.local/bin/bud-decision-studio.AppImagerm ~/.local/share/applications/bud-decision-studio.desktop

Running from source instead#

Developers can run the studio from a checkout of the repository, without the desktop app:

install.sh detects your hardware, asks where models run and installs everything into .venv, exactly as setup does in the app. A source checkout keeps its data in ./data instead of the app's folder. Run from source has the details.

Terminal
Shell
git clone https://github.com/BudEcosystem/Bud-Decision-Engine.git && cd Bud-Decision-Engine./install.sh      # Windows: powershell -ExecutionPolicy Bypass -File install.ps1./run.sh          # then open http://127.0.0.1:8420