Privacy Policy

Audio recordings and transcripts

When a Game Master chooses to record or upload session audio, we temporarily store it while transcription is pending or processing. AssemblyAI converts the recording to a text transcript. After processing, we attempt to delete the source audio and request deletion of AssemblyAI's copy. Pending jobs and cleanup failures can delay deletion. We do not retain completed recordings as world audio assets. The resulting transcript is saved to the session and treated like other text you create in Storywright.

• What we store: audio temporarily during processing, the transcript text (and optional timing segments), the session it belongs to, and processing metadata (duration, timestamps).
• Where audio is processed: a temporary object-storage location and AssemblyAI, the speech-to-text provider named below. After processing, we attempt to delete the temporary source audio and request deletion of the provider's copy of the audio and transcript through its API.
• Retention: temporary audio remains while transcription is pending or processing, and cleanup failures can delay deletion. Transcripts and text drafts follow the applicable account and world deletion processes. Content in a transferred world may continue with its new owner.
• Training: we do not use recordings or transcripts to train AI models.

Sub-processors for audio and transcripts

• AssemblyAI (speech-to-text), which processes the audio to return text. After processing, we request deletion of its copy of the audio and transcript. The account is configured with AssemblyAI's no-training and zero-retention options.
• Anthropic (text drafting), which processes transcript and session text — never audio — when a GM or player asks Storywright to draft a recap, journal entry, or similar text. Anthropic's API terms do not permit training on our data.