Data Manager
The Data Manager is the tabular view of every task in a project. It's where annotators pick tasks to work on and managers do bulk operations.
Click any project → Data Manager in the tab strip (this is the default landing page when you open a project).
Split a recording into clips
Tick one task that holds audio or video and press Split in the
toolbar (project managers only, in projects with an audio or video
field). The recording opens in the splitter: play it, drop cut points,
or let silence detection suggest them, then cut. The clips become new
tasks with clip_index, clip_start, clip_end and source_file
fields; the original task stays until you delete it from the summary
screen. Full guide: Splitting recordings into clips.
What's on the page
Three regions:
- Header bar at the top — view tabs, filters, ordering, column picker, Label All Tasks, export / import, count
- Table in the middle — one row per task, configurable columns
- Selection footer at the bottom — appears when rows are selected; bulk actions
Columns
Shown by default:
| Column | What it shows |
|---|---|
| ID | Task ID — clickable, opens task preview |
| Completed | When the task reached its annotation target (is_labeled) |
| Annotations | Count of human annotations on this task |
| Cancelled | Count of skipped (cancelled) annotations |
| Predictions | Count of model-produced pre-annotations |
| Annotated by | Who submitted annotations |
Available from the column picker:
| Column | What it shows |
|---|---|
| Inner ID | Per-project sequential task number |
| Annotation results / Annotation IDs | Raw result JSON / IDs |
| Prediction score / Prediction model versions / Prediction results | Prediction details |
| Upload filename / Storage filename | Where the task came from |
| Created at / Updated at / Updated by | Task timestamps and last editor |
| Lead Time | Average annotation lead time |
| Drafts | Whether a draft exists |
| Heavy file | Whether the task's media file is long or large enough to load slowly. Empty until Measure files has run |
| File size | Size of the task's media file, in bytes. Empty until Measure files has run |
| Length | Length of the task's recording or video, in seconds. Empty until Measure files has run |
Plus one dynamic column per data field in the project's task
schema — for an Image+Text project you'll see image and text
columns surfacing the actual values.
Configure columns
Click the Columns control in the header → toggle visibility of any column. Drag column headers to reorder. The layout is saved on the tab.
Tabs (saved views)
A tab is a saved combination of filters + sort + columns. Useful for recurring queries like "all tasks I've skipped" or "unlabelled tasks".
Default tab: Default — every task, default columns, no filter.
Create a tab
- Click + in the tab strip
- Apply the filters / sort / columns you want
- Rename it (double-click the tab title)
Tabs are saved on the project; there is no private / shared toggle. The URL carries the tab ID, so a shared link opens the same view for a teammate with access to the project.
Filters
Click Filters to add one. Each filter has:
- Column — which column to filter on
- Operator — comparison (equals, contains, between, before, after, is empty, is not empty, etc.)
- Value(s) — the comparison value(s)
Filters combine with And or Or (conjunction picker at the top of the filter list).
Common filters:
- Annotations = 0 — unlabelled tasks
- Completed is empty — not yet at the annotation target
- Cancelled > 0 — skipped at least once
- Updated at > 7 days ago — stale tasks
- Predictions > 0 — tasks with pre-annotations
Sort
Pick a column in the Order control (or click a column header); pick again to reverse. Sort + filter both persist on the tab, and the URL identifies the tab, so reloading keeps your view intact.
Bulk actions
Select rows via the leftmost checkbox column (or check the header checkbox to select all visible rows). The selection footer appears at the bottom with these actions:
| Action | What it does |
|---|---|
| Label N tasks | Open the labelling stream pre-filtered to these tasks |
| Retrieve Predictions | Ask the project's ML backend to predict the selected tasks |
| Create Annotations From Predictions | Copy the selected tasks' predictions into annotations (pick the model version) |
| Delete Annotations | Delete every annotation on selected tasks. Tasks stay. Confirmation modal |
| Delete Predictions | Delete every prediction on selected tasks |
| Delete Tasks | Delete the tasks themselves. Annotations cascade. Confirmation modal |
| Cache Labels | Write the tasks' labels into a data field (for filtering / export) |
| Remove Duplicated Tasks | Drop tasks whose data duplicates another task |
| Measure files | Read the size and length of each selected task's media file (from a bucket, an upload or a web address) into the Heavy file, File size and Length columns. Large selections are measured a part at a time: what is left is measured overnight, or run it again to do more now. What counts as heavy is the organisation's policy |
Actions appear only for roles with the matching permission (e.g.
deleting tasks needs tasks_delete).
Note: "Select all visible" only picks tasks in the current table page. To act on more, use Select all in the header — it selects every task matching the current filter, even ones not on screen.
Export (header bar) snapshots the current tab's tasks to a downloadable file in your chosen format. See Data export for formats.
Task preview
Click a task's ID (or row, anywhere outside a checkbox) to open the task preview in a side panel. The preview shows:
- The task's
data - Every annotation on the task (tabs at the top, one per annotation)
- Every prediction (model-produced pre-annotations)
- Comments thread
The preview is the operator's way to spot-check a task without opening the full labelling stream.
Scope
The Data Manager is per-project. Workspace landing pages list projects but don't aggregate their tasks; the per-project Data Manager is the destination for task-level work.
Performance
Data Manager pages a limited slice of tasks at a time (default 30 per
page). Large projects work fine — filtering + sorting hit indexed
columns. Pagination + a high page_size query parameter let you walk
the full dataset programmatically.
API
The Data Manager UI is a thin shell over the REST API. Anything you can do in the UI you can do via the API:
# List tasks
curl "{{platformURL}}/api/projects/<id>/tasks/?page=1&page_size=100" \
-H "Authorization: Bearer YOUR_PAT"
# Bulk-delete annotations on selected tasks (same endpoint the footer uses)
curl -X POST "{{platformURL}}/api/dm/actions?id=delete_tasks_annotations&project=<id>" \
-H "Authorization: Bearer YOUR_PAT" \
-H "Content-Type: application/json" \
-d '{"selectedItems": {"all": false, "included": [101, 102]}}'
GET /api/dm/actions?project=<id> lists the action IDs available to
you. See API → Overview for the full endpoint list.
What's next
- Labeling guide → — what happens when you click Label All Tasks
- Data export → — exporting annotations out
- Review workflow → — the reviewer side (assignment lives in Settings → Review)