Files & Models

This page documents the desktop file formats and local save behavior.

File types

.dvp

The dAIve project file (a zip archive internally).

It stores the complete project workspace, including:

• configuration and workflow state
• training data references
• trained model files
• analysis results
• predictions

This is the main file you reopen when continuing work later.

.dvm

The exported trained model file.

Use it when you want to run prediction without carrying the full project.

It is the right format for Only Prediction workflows.

A project can contain multiple trained models internally (e.g. from batch training), but the .dvm export always contains the currently active model.

.csv

Desktop CSV files are used for:

• training data
• manual validation and test data (when using manual data splits)
• prediction input data
• exported prediction tables

Local save behavior

Desktop projects are stored on the local filesystem.

Important behavior:

• New Project creates and saves the project immediately at the chosen path
• Open Project restores the saved local project
• Save & Close Project writes the current state back to the same .dvp
• Save project as... (from the Project menu) saves a copy under a new name and path

If you close a project with unsaved changes, the app prompts:

• Save and close — saves first, then closes
• Discard and close — closes without saving
• Cancel — returns to the project

Desktop saving is file-based, so path choice and local file management matter more than in the web app.

Model export

After successful training, Model Training can export the current model as a .dvm file.

Project artifacts

Inside the .dvp archive, dAIve manages several generated artifacts:

• trained model files (.dvm)
• data distribution analysis (data_distribution.json)
• data correlation analysis (data_correlation.json)
• data size analysis (data_size_analysis.json)
• input dropout analysis (input_dropout_analysis.json)
• predictions (predictions.csv)
• project settings (projectSettings.txt)

These are managed automatically. Changing inputs, outputs, or retraining invalidates the relevant artifacts. The Clear Invalid Data action in the Data menu can also remove them manually.

No cloud project list

Unlike the web app, the desktop product does not manage projects through a cloud project overview with card actions such as Rename, Delete, or Export.

That means file organization, backup copies, and naming conventions are handled through the operating system and local filesystem.