Gen3 Datasets Managing

Managing Dataset Content

Start Here

1. Open Datasets and open a dataset's documents view.
2. Review the documents table for status, preprocessing stages, errors, and duplicates.
3. Use View summary to open the floating document summary panel for a selected file.
4. Remove or replace stale files before major reviews; use bulk reprocess/delete when many rows need the same action.
5. Re-share or adjust group access when membership changes.

Why this matters

Managing content keeps retrieval trustworthy—operators trust answers only when the underlying document set is current.

Details

Dataset management in active Gen 3 stays inside the datasets workspace. Use the dataset documents view and the dataset edit form together to keep the retrieval set accurate, current, and appropriately shared.

Typical management tasks

• review which documents are in the dataset
• remove outdated files
• upload replacement files
• confirm import results
• review sharing posture
• verify retrieval defaults such as embedding model and chunking strategy

Document-level review

Open the dataset documents view when you need to inspect the files inside one dataset. This is the right place to confirm whether ingestion succeeded and whether old material should be removed.

Bulk document actions

When you can mutate the dataset, select rows (or all visible rows) to run bulk Reprocess or Delete from the documents modal bulk bar. Dataset-level ZIP export remains on the Datasets hub—not per-document JSON export.

Floating document summary panel

View summary on a document row opens a draggable, resizable floating panel with the generated summary text. Use it to skim long corpora without leaving the documents table.

Preprocessing progress stages

While ingestion runs, the progress column shows human-readable stages such as Queued for processing, Preparing document, Reading document, Analyzing images, Transcribing audio, Generating embeddings, Generating summary, and Finalizing. Wait for Ready (or resolve Processing failed) before relying on retrieval in GT Chat.

Dataset-level review

Return to the dataset edit workflow when you need to change:

• name or description
• access posture
• group sharing
• retrieval defaults
• other metadata that affects how the dataset should be used

Management workflow

1. Open the dataset.
2. Review the current document list.
3. Remove outdated or incorrect material.
4. Upload or import replacement content.
5. Re-check the sharing posture.
6. Validate the result in GT Chat or in any agent that depends on the dataset.

When to edit the dataset instead of the agent

If the problem is bad or missing source material, fix the dataset. If the problem is how the agent uses otherwise-correct source material, then review the agent instead. Separating those two causes prevents unnecessary agent churn.

Best practices

• Keep one dataset aligned to one coherent body of source material.
• Remove stale files promptly so retrieval does not mix old and new guidance.
• Revisit group sharing after imports, because imported content may deserve a narrower or broader audience.
• Test the real downstream workflow after major changes.

Related pages

• Datasets
• Uploading and Importing Documents
• Agents