Gen3 Datasets - GT-Edge-AI/GT-AI-OS GitHub Wiki

Datasets

Start Here

  1. Open Datasets from the tenant sidebar — use quick view for favorite tiles or detailed view for the full hub.
  2. Create a dataset for each logical corpus (project, unit, assessment cycle); pick, create (+ Create new category…), edit, or delete categories inline when editing metadata (creators manage their categories; tenant owners manage any).
  3. Upload or import documents (Uploading and Importing Documents).
  4. Attach datasets in chat or agent defaults when you are ready to query them.

Why this matters

Datasets are the durable knowledge layer for chat retrieval, workflow review, and GT API upload automation.

Details

Datasets are the active Gen 3 retrieval boundary for tenant documents. The current Datasets page owns dataset creation, document upload/import, ZIP import/export, document review, sharing posture, and deletion. Gen 3 no longer relies on a separate tenant Documents route for normal document operations.

What you do on the Datasets page

  • create new datasets
  • edit dataset metadata and retrieval defaults
  • upload files directly into a dataset
  • import datasets from ZIP bundles or legacy document bundles
  • export one or more datasets
  • open the per-dataset document view
  • share datasets privately, to groups, or more broadly when allowed by the deployment

How the page is organized

Quick view and detailed view

Mirroring Agents, Datasets exposes:

  • Quick view (?view=quick) — favorite dataset tiles, Add Favorites, Custom sort with drag reorder
  • Detailed view (?view=detailed) — full hub with create/import/export/delete actions

Sidebar Datasets submenu links to the same two views.

Dataset list and filters

The shared agent-style filter toolbar lets you narrow the list by search, category, tags, access (All / Mine / Organization), and sort order. Filter and sort choices persist per browser so returning operators keep their last posture.

Bulk actions

Datasets supports bulk export, bulk delete, and bulk upload entry points so you do not have to open one dataset at a time for every operation.

Per-dataset document workspace

Opening a dataset's document modal lets you inspect the current document list, upload more files, import additional documents, and remove outdated content.

Common tasks

Create a dataset

  1. Open Datasets (detailed view).
  2. Select Create Dataset.
  3. Enter the dataset name and optional description.
  4. Choose a category from the catalog, select + Create new category… to add one inline, or use Edit category / Delete category when your role allows.
  5. Choose the access posture and any required group shares.
  6. Save the dataset.

Add source material

Use direct upload when you already know the destination dataset. Use import when you are bringing in a packaged dataset archive or a larger prepared file set.

Review dataset content

Open the dataset's documents view to confirm the files present, the ingestion results, and whether outdated material should be removed or replaced.

Datasets and chat

Datasets can be used in two places:

  • as default datasets attached to an agent
  • as conversation-level datasets attached inside GT Chat

Use agent defaults for repeatable workflows and conversation-level attachments for one-off context.

Sharing model

Dataset visibility depends on the access option selected for the dataset and, when group-shared, on the relevant group. If another user cannot see a dataset, check its sharing posture before assuming the upload failed.

Best practices

  • Use clear dataset names that match a team, project, or content purpose.
  • Remove outdated documents instead of leaving conflicting versions in the same retrieval set.
  • Prefer group sharing over broad visibility when the content is team-specific.
  • Revisit retrieval defaults after import so the dataset uses the intended embedding and chunking posture.

Supporting pages