Engineering Workspace

Data Assets

Curate datasets from agent traces, uploads, and agent-produced artifacts. Data assets feed evals, SFT, preference tuning, and Lab-ready snapshots.

CLI

stimulir data wraps the endpoints below. The most common flow is from-trace → stage → snapshot — turn an agent run into staged data, then pin a snapshot the Lab consumes for evaluation, SFT, or preference tuning.

bash
# List data assets in the active workspace
stimulir data list
stimulir data list --target eval

# Bring your own file in as a new asset
stimulir data upload ./dataset.jsonl --stage raw --target eval

# Curate an asset directly from a recorded trace
stimulir data from-trace <trace-id> --source agent --target eval

# Promote through the curation stages: raw → cleaning → clean_view → snapshot → lab
stimulir data stage <asset-id> clean_view
stimulir data stage <asset-id> lab

# Move an asset back to raw or stage several assets together
stimulir data unstage <asset-id>
stimulir data bulk-stage --ids <asset-id>,<asset-id> --stage lab --target eval

# Update hub metadata without changing the stored payload
stimulir data update <asset-id> --name "Clean eval view" --target eval

# Pin an immutable snapshot a Lab eval or training run can consume
stimulir data snapshot <asset-id>

# Remove an asset (asks for confirmation; --yes skips it)
stimulir data remove <asset-id>

Endpoints

GET/api/v1/workspace/data-assets
POST/api/v1/workspace/data-assets
POST/api/v1/workspace/data-assets/uploads
POST/api/v1/workspace/data-assets/from-trace
POST/api/v1/workspace/data-assets/bulk-stage
PATCH/api/v1/workspace/data-assets/{asset_id}
PUT/api/v1/workspace/data-assets/{asset_id}/stage
POST/api/v1/workspace/data-assets/{asset_id}/unstage
POST/api/v1/workspace/data-assets/{asset_id}/snapshot
DELETE/api/v1/workspace/data-assets/{asset_id}

Platform endpoints authenticate with your session token and the X-Business-Profile-Id header.

List & create

GET/api/v1/workspace/data-assets
POST/api/v1/workspace/data-assets

List the data assets in the active workspace, or create a new one. Use query filters such as stage=lab, status=ready, and target=eval to load the assets that should appear in the Lab eval hub.

Sources

Assets are populated from two sources:

POST/api/v1/workspace/data-assets/uploads
POST/api/v1/workspace/data-assets/from-trace
  • Uploads — bring your own files into a data asset.
  • From trace — turn a recorded agent trace into curated data. This is how Engineering sessions — including agent runs — feed back into training data.

Stage for Lab

PATCH/api/v1/workspace/data-assets/{asset_id}
PUT/api/v1/workspace/data-assets/{asset_id}/stage
POST/api/v1/workspace/data-assets/bulk-stage
POST/api/v1/workspace/data-assets/{asset_id}/unstage

Update metadata, move one asset or many assets through curation stages, or unstage an asset back to raw. Targets distinguish how the asset will be used: eval, sft, or preference. See Lab Evaluation for durable prompt/data eval runs.

Snapshot & delete

POST/api/v1/workspace/data-assets/{asset_id}/snapshot
DELETE/api/v1/workspace/data-assets/{asset_id}

Take an immutable snapshot of an asset to pin the exact data an eval or training run used, or delete an asset you no longer need.