Studio Tour

One workspace, many modules

The Studio is a single page with an icon sidebar on the left. Each icon opens a module, and every module you have opened stays loaded in the background, so switching away and back does not lose open files, loaded datasets or a half-built chart.

The sidebar, top to bottom

• Projects: create, open, rename and delete projects. See Projects.
• Explorer: the Code Editor. While it is open the sidebar adds three tools of its own: Search across files, Source Control and the AI Copilot.
• Model Builder: design a network on a canvas and train it. See Visual Model Builder.
• Data Hub: your dataset library. While it is open the sidebar adds Database connections and PrepFlow. See Data Hub and PrepFlow.
• Pipelines: repeatable, schedulable runs. See Pipelines.
• Research: turn a paper into a runnable project. See Research.
• Dashboard: charts, parameters and stories. See Dashboard.
• Deployment: try a model and export it. See Deployment.
• Fine-Tune: continue training a model or write a LoRA script. See Fine-Tuning.
• Settings and account: appearance, compute, AI provider, spending limits and your account. See Settings and account.

An Extensions marketplace exists in the code base but is switched off in the sidebar.

How the modules hand work to each other

Most of the Studio is about passing an artifact from one module to the next without exporting and re-importing it:

• A dataset loaded in the Data Hub is the active dataset for PrepFlow, the Model Builder, the Dashboard and the Code Editor.
• A PrepFlow graph can be materialized back into the Data Hub as a new dataset version, or reused as a step in a Pipeline.
• A model trained in the Model Builder is kept in the project and can be opened in Fine-Tune, Deployment or a Pipeline from the training run's summary.
• A pipeline can refresh the Dashboard, build a deployment package and open it in Deployment.
• Research generates ordinary project files, a Model Builder graph and Dashboard figures, so everything it makes is editable in the module that normally owns it.

The Project page can draw these relationships for you: its Connections card lists, for each item, what it was built from and what was built from it.

Tools available everywhere

• Command palettes: the Code Editor has one (Ctrl+Shift+P, or ⌘⇧P on a Mac) and the Data Hub has its own (Ctrl+K). The Model Builder opens a layer search with Ctrl+K. See Keyboard shortcuts.
• Job Centre: imports, profiling, conversions, pipeline runs and Research figures run as background jobs. See Jobs and notifications.
• Status bar: cursor position, language, indentation, the current git branch and the editor's error and warning counts.
• Resource monitor: live CPU, memory and GPU usage while a model trains, and the Model Builder's memory estimate.
• Source control: a git client that works offline, on the same files as the editor.

Where your work is saved

A project is saved in your browser's own storage, or, in Chrome, Edge and other Chromium browsers, in a real folder on your disk. Either way the project's asset list is mirrored to a manifest, so a refresh does not lose anything. See Projects and Privacy and storage.

Where to go next

Follow the Quick Start or work through the tutorials, starting with Train a Classifier in Your Browser.