Data Export
On a deployed box, upload run data that skills mark for export to an S3 bucket your lab owns.
Data Export points a deployed box at an S3 bucket you control. Skills then mark individual captures, API responses or whole runs for export, and the box uploads them to that bucket. Nothing is exported unless a skill asks for it.
Data Export exists on a deployed box only. The cloud app has no Data Export entry. A skill that asks for export where no bucket is configured exports nothing and logs a warning.
Before you start
- The bucket must already exist. Nothing creates it.
- Use an access key limited to that bucket. The dialog states the minimal policy:
s3:PutObject, pluss3:DeleteObjectfor the connection test, onarn:aws:s3:::<bucket>/<prefix>/*. The key is stored on the box and used for every export anyone triggers there.
Configuring it
Open the header menu (the vertical dots) and choose Data Export. The dialog is titled Data Export — S3 destination.
| Field | Notes |
|---|---|
| Bucket | The bucket name. |
| Region | For example us-east-1. |
| Prefix (optional) | A path to put everything under, for example zeon/robot-1. |
| Access key ID | |
| Secret access key | |
| Session token (optional) | For temporary credentials. |
Test connection writes and deletes a small test object, so a wrong bucket or key shows now rather than after a run. Save stores the configuration.
Credentials are never sent back to the browser. After saving, the dialog shows AWS key •••• and the last four characters of the key ID (and + session token when one is stored). Replace credentials… asks for a new key ID and secret together; Keep current backs out. You can change the bucket, region or prefix without re-entering the credentials.
Once saved, the dialog also shows Configured by (who saved it, and when), Uploads pending, and, while uploads are failing, Last upload error. Remove connection deletes the bucket settings and credentials from the box.
Anyone signed in on the box can open the dialog and run Test connection. Once a connection exists, only the person who configured it and admins (organization admins and Zeon administrators) can change or remove it.
What gets exported
| Skill call | What is uploaded | Where it lands |
|---|---|---|
capture_image(..., export=True) | That capture | <prefix>/<project>/<run-id>/captures/<capture-name>/ |
api_request(..., save_name=..., export=True) | That response | <prefix>/<project>/<run-id>/api/<save-name>.json |
print_log(..., export=True) | The whole run folder, once the workflow run ends | <prefix>/<project>/<run-id>/run/ |
<project> is the project's name, or its id when it has no name. One print_log(..., export=True) anywhere in a run is enough to export that run's folder. The calls are described in the skill runtime API.
While a destination is configured, a workflow run's log starts with an exports → … line naming it.
When an upload fails
Uploads run in the background and never slow down or fail a run. Each file is tried three times. After that the error shows as Last upload error in the dialog until an upload succeeds again. Export never deletes files on the box, so a failed upload loses nothing there.
A capture's depth data arrives compressed, as depth.npz instead of depth.npy.
Updated 9 days ago