Datasets Sharing

Gen 3: This is a legacy Gen 2 article. For current GT AI OS 3.0 guidance, see gen3/datasets/managing.

Sharing Datasets

This guide explains how to share datasets with team members, across your organization, and how to export/import datasets.

Visibility Levels

Visibility	Who Can Access	Best For
Individual	Only you	Personal knowledge bases
Team	Selected team members	Team-specific resources
Organization	All users in your organization	Company-wide knowledge

Note: Only Tenant Admins can set visibility to Organization. All users can access Organization-level datasets.

Setting Visibility

Changing Dataset Visibility

1. Open the dataset details page
2. Click Edit or access settings
3. Change the visibility setting
4. Select teams (if Team visibility)
5. Save changes

Sharing with Teams

Team Access

When you share with a team:

• Team members see the dataset in their list
• They can use it with their agents
• They can view document contents

Selecting Teams

1. Set visibility to Team
2. Select which teams should have access
3. Save changes

Revoking Access

1. Edit the dataset
2. Deselect teams or change visibility to Individual
3. Save changes

Organization-Wide Sharing

Making Datasets Organization-Wide

1. Edit the dataset
2. Set visibility to Organization
3. Save changes

All organization members can now access the dataset.

Considerations

Before sharing organization-wide:

• Ensure content is appropriate for all users
• Verify data sensitivity is acceptable
• Document what the dataset contains
• Consider maintenance responsibilities

Export and Import

Exporting Datasets

Export datasets for backup or transfer to another GT AI OS instance:

1. Go to the Datasets page
2. Select one or more datasets using checkboxes
3. Click Export
4. A ZIP file downloads

Export contents:

• metadata.json: Dataset name, description, category, tags, and settings
• documents.json: List of uploaded documents with their metadata
• chunks.jsonl: Processed text chunks used for search

Note: Maximum export size is 100MB.

Importing Datasets

Import previously exported datasets:

1. Go to the Datasets page
2. Click Import
3. Select the ZIP file
4. Review the configuration and any warnings
5. Choose whether to re-process chunks
6. Click Import

Re-processing chunks: When moving datasets between different GT AI OS instances or versions, it's recommended to re-process the chunks. This ensures optimal search quality with the current embedding model. If you're restoring a backup to the same instance, you can skip re-processing.

Export and Delete Operations

Export and delete actions work for both individual datasets and bulk selections:

Single dataset:

• Click on a dataset, then use the Export or Delete button

Multiple datasets:

• Select datasets using checkboxes on dataset cards
• Click Export to download all selected as a ZIP
• Click Delete to remove all selected (with confirmation)

Best Practices

Before Sharing

1. Review content: Ensure accuracy and appropriateness
2. Document purpose: Write clear descriptions
3. Set proper visibility: Don't over-share
4. Add relevant tags: Help others find the dataset

Managing Shared Datasets

1. Communicate changes: Let users know about updates
2. Version carefully: Major changes may affect agents using the dataset
3. Monitor usage: Check if dataset is being used effectively
4. Delete when outdated: Remove stale content

Security Considerations

• Don't share sensitive information inappropriately
• Review content before organization-wide sharing
• Consider access controls for sensitive data
• Follow your organization's data policies

Troubleshooting

Users Can't See Shared Dataset

• Verify visibility settings are correct
• Confirm users are in the selected team(s)
• Check that the dataset is active

Access Permission Issues

• Review your sharing settings
• Ensure you have permission to share
• Contact your Tenant Admin if needed

Import Warnings

If you see warnings during import:

• Missing embedding model: Re-process chunks recommended
• Large file size: May take longer to process
• Duplicate names: Consider renaming before import