Design or open
Generate a PyTorch architecture with the local assistant, or open an existing Python model script and continue from there.
NIXTEE DESKTOP · STANDALONE MODEL LAB
From model idea to training run, in one workspace.
Design, inspect, validate and train PyTorch models in a dedicated desktop app. Your files, datasets and editor live together—no terminal workflow required.
WindowsmacOSLinux x64Local workspace
01 DESIGN02 SEARCH03 TRAIN04 SHIP
Design a neural network CSV · prices.csv
DESIGN BRIEF
Create a compact model to predict the next-hour BTC close from recent price and volume data.
◆ model.pyPYTHON
class Net(nn.Module): def __init__(self): super().__init__() self.layers = nn.Sequential( nn.Linear(8, 64), nn.ReLU(), nn.Linear(64, 1) )
Architecture readyValidate architecture →
Workspace files stay on this deviceUTF-8 · Python
ONE DESKTOP WORKSPACE
Everything around your model, in one place.
Move from a design brief to a saved model and training metrics without stitching together terminal commands and separate tools.
Browse and edit
Manage workspace files and CSV datasets, then edit Python and data files in the built-in tabbed editor.
Validate and search
Optionally score an architecture with Nixtee GNN. Search can refine candidates in a loop, with or without a separate validation step.
Train and review
Continue to local training, follow per-epoch metrics and keep the resulting model checkpoint in your workspace.
YOUR FLOW, YOUR CHOICE
Use GNN scoring when it helps. Keep moving when it doesn’t.
Nixtee brings architecture validation, iterative Search and training into the same desktop flow. You can validate first, start Search directly, or continue straight to training.
See the desktop quick startValidate architectureOptional GNN performance estimate
OPTIONALSearch candidatesIterate and compare architecture scores
GNN LOOPContinue to trainingTrain locally and inspect epoch metrics
LOCAL◉
Local work, connected scoring. Model design, workspace files, datasets and training run in the desktop app. GNN validation and Search scoring connect to the Nixtee API and use your account credentials.
NIXTEE DESKTOP · RELEASE 1.0.5Installation guide →
Download Nixtee for your computer
Choose your operating system and processor. Includes secure MCP connections, per-tool approvals and OAuth. Local components are installed on first launch.
Python CLI 0.9.13: Wheel · Source archive · SHA-256 checksums
MCP model predictions require Docker and the offline sandbox image. MCP setup and migration →
The macOS release is currently unsigned and has not been notarized by Apple.
START WITH NIXTEE DESKTOP
Install the app. Choose a workspace. Build your model.
The first launch guides you through setting up the local AI and Python components the app needs. Create a Nixtee account to get API credentials for GNN validation and Search.
01 Install the Windows, macOS or Linux app02 Set up local components on first launch03 Add your Nixtee API credentials when ready
Read the desktop setup guide NIXTEE ACCOUNT
Create your account
Use your account credentials in the desktop app when you want GNN validation and Search scoring.