Managing projects

A project in Ertas is a workspace inside the Model Studio tab. Each project has its own canvas state, its own training-run history, and its own name. Datasets and model artifacts live outside projects (in Data Craft and Hub respectively) and are accessible from any project.

This page covers the mechanics: how to navigate between projects, how to organise the canvas as it grows, and how to clean up.

What lives in a project (and what does not)

This separation matters when you delete things: see the table at the bottom of this page.

Navigating between projects

Two ways to switch projects:

Ertas restores your previous session on sign-in. Whatever page you had open when you logged out (a specific project's canvas, the Runs tab, Data Craft, anywhere) is what loads next time. You do not have to navigate back to where you were.

A useful organising habit is one project per model you are actively building. A customer-support fine-tune and a code-completion fine-tune deserve separate projects so their canvases do not clutter each other.

Creating, renaming, and deleting projects

From the Model Studio home, you can create a new project (empty canvas) or open an existing one. Inside a project, the breadcrumb and project menu let you rename or delete.

Deletion is immediate and irreversible. It removes:

• The canvas state (recipes, node positions, sticky notes).
• The project name and metadata.

It does not remove:

• Datasets in Data Craft (those are account-wide).
• Hub entries (LoRA + GGUF) produced by runs in this project. Those persist in Hub until you delete them manually.
• Run records in the Runs tab. Those also persist until manually deleted.

There is no archive option. If you might come back to a project, leave it in place; deleting and recreating it loses the canvas geometry.

Keeping the canvas tidy

A canvas with one recipe is easy. A canvas with ten is a UX problem. A few habits that keep things readable:

• Group related recipes spatially. Studio does not enforce groups, but using the canvas geometry (one cluster per experiment thread) makes the workspace easier to read.
• Delete child nodes that are not connected. If you abandoned an experiment halfway, the unused base model or dataset nodes still take up real estate. Select and delete.
• Use descriptive recipe names. The hover label on an Action Module is the first thing you see. Support v3, rank 32 is more useful than Fine-Tune.
• Sticky notes for team context. Pin a sticky note next to an Action Module ("v3 overfits, try rank 8 next") so whoever opens the project next has the rationale.
• Delete recipes you are done with from the canvas. The run history in the Runs tab is preserved separately, so any past run's config and artifacts are still reachable from there.

You can pan and zoom the canvas with mouse and trackpad. The keyboard shortcuts dialog (press ? on the canvas) lists the navigation controls.

Run history

Training runs live in the dedicated Runs tab (a root tab in the navigation, not inside a project). The tab shows every run in your account, grouped into Active (queued, provisioning, training, retrying) and History (completed, failed, cancelled). The only filter is by status; sort order is descending by start time.

The build canvas inside Model Studio also has a Run panel that surfaces active runs related to the modules currently visible. Both views read from the same underlying records.

Deleting a run record from the Runs tab removes the run from the list. It does not remove the model files (LoRA + GGUF) produced by that run. Those live in Hub and must be deleted there separately.

Collaboration

Multi-account real-time collaboration is not yet a feature. Teams that need shared visibility today typically do one of two things:

• Share a single account with the team. The canvas supports sticky notes, which are a lightweight way to leave context for whoever opens the project next.
• Trade screenshots and downloaded artifacts for review. Every run produces a downloadable GGUF and LoRA from Hub, so teammates can evaluate without needing canvas access.

Storage and quotas

Datasets and model artifacts are billed against account-wide quotas, not per-project quotas. Plan limits:

See Storage for the full cleanup walkthrough.

What gets deleted when

A quick reference for the three independent delete actions:

If you want to fully clean up everything associated with one experiment thread, you need to delete in all three places: Runs tab, Hub, and the canvas recipe.

What's next