Core Guides
Data retention
How long you can access request details and media, and how to keep important results
Request and response details, and automatically saved media results, are available for 30 days. Conversation history, billing records, saved Media Assets, and files you download are handled separately.
See TokenLab data retention for the full explanation.
Request and response details
Saved request and response content is available in Console, Request Debugger, and TokenLab Agent (Assistant) for 30 days from when it is recorded. After expiry, these tools can no longer retrieve that original content. Request summaries, usage, and billing records may still be available. Download any details you need before they expire.
An expired capture is different from content that was never saved or could not be loaded. Check the displayed status when troubleshooting.
Generated media
TokenLab automatically saves copies of supported image, video, audio, and downloadable 3D results (including Worlds files) returned as HTTP(S) links. A successfully saved copy is available for 30 days from when the result is first recorded. Viewing it again, retrying, or checking task status does not restart that period.
Check the result's saving status and expiry. If saving is still in progress or has failed, the original link may stop working before a copy is available. For longer access, download the file or save it to Media Assets while it is available. A copy saved to Media Assets does not expire with the automatic 30-day copy.
Automatic saving does not cover interactive Worlds pages, externally referenced model dependencies, inline Base64 or binary media, input files, or links within generated text. Save these results yourself if you need them later. Task records have a separate lifetime; a completed task does not guarantee its media is still available.
Checking media availability through the API
AI-protocol JSON responses include media_retention for supported results: days and items, with each item reporting url, kind, status, and expires_at. You can also add metadata=true to a media-result URL to check its current status. A saving or failed status does not mean the file has been saved. An expired result URL returns HTTP 410.
A download URL obtained from an available result lasts at most five minutes. If it expires while the result is still within its 30-day window, request a fresh URL. This does not extend the result's expiry.
TokenLab Agent (Assistant) conversation history
These are your conversations with the built-in TokenLab assistant. They are stored separately from API request logs and are not automatically removed when those logs reach 30 days. Text already saved in a conversation may remain there, but expired request details and media cannot be loaded again from that conversation.
You can remove a conversation from Agent history. This removes it from the history list; it does not erase all associated data or remove request, billing, support, or Console history. The Assistant can investigate original request and response content only while it is within the 30-day window; older billing and usage records may still be available.
Other records and files
- Account information, billing and usage history, and task records are kept separately to provide account services, billing, and support. Their retention is not limited to the 30-day request-detail window.
- Organization call-record export files are available to download for 90 days. Download them before they expire; expiry of an export does not delete the underlying records.
- Temporary files uploaded in Console or TokenLab Agent (Assistant) can be used in new requests for 24 hours. After that, upload the file again if you need to use it in a new request.
- Files you download or export, browser storage, and copies saved by your own application are under your control. TokenLab's retention limits do not delete those copies.
The periods above describe availability through TokenLab. Deletion from storage may finish later than access expiry. AI service providers that process your requests have their own data policies and retention periods.
For request troubleshooting, see Logs and troubleshooting. For media task polling, see Async jobs and polling.