PrepBoss handling of customer intellectual property
Assessment date: August 29, 2026
Summary
- Customers retain ownership of manuscripts, notes, and voice references under the PrepBoss Terms. PrepBoss receives a limited license to process them to operate the service.
- The PrepBoss application contains no model-training pipeline. PrepBoss's Terms and Privacy Policy state that customer content is not used to train or evaluate internal models.
- OpenAI is the application's only external AI model provider. Manuscript text and related analysis context are sent to OpenAI; across processing calls, this can include most or all of the manuscript.
- OpenAI is not permitted to use data received from PrepBoss for model training.
- Recorded voice samples are stored for playback and reference. They are not sent to OpenAI or another speech-model API, and the code contains no voice-cloning or speaker-biometric pipeline.
- Customer notes are stored in MongoDB Atlas and are not sent to OpenAI.
- Book narration audio is never submitted to PrepBoss.
- Stored customer content is encrypted at rest by the storage providers.
- No data is shared with any other services for any purpose.
Data flows
Manuscript
Notes
Voice samples
Recorded voice audio has no current path to OpenAI or another speech-model API.
Services that receive customer content
The application runs on Fly.io, where customer content passes through the backend during processing. DigitalOcean Spaces and MongoDB Atlas are the application's persistent customer-content stores.
| Service | Manuscript | Notes | Voice samples | Use |
|---|---|---|---|---|
| DigitalOcean Spaces | Original file, extracted text, and full OpenAI request/response diagnostics | — | WAV audio objects | Primary object storage with provider-managed encryption at rest |
| MongoDB Atlas | Structured and derived book data, excerpts, character data, dialogue spans, and vocabulary | Character notes, vocabulary notes, and bookmark notes | Sample IDs and descriptive tags, not audio bytes | Primary application database with provider-managed encryption at rest |
| OpenAI API | Text chunks, dialogue or narrative excerpts, phrases, and derived character/scene context | — | — | Manuscript analysis and text-to-speech pronunciation |
| Forvo / Merriam-Webster | One selected word or short phrase; Merriam-Webster does not receive the surrounding excerpt | No | No customer audio | Dictionary and pronunciation lookup |
| New Relic | Occasional words and character names, retained for seven days | — | — | Logs and request tracing |
Evidence regarding AI training by third-party service providers
OpenAI's current Services Agreement states that the customer retains ownership of input, owns output, and that OpenAI does not use customer content to develop or improve its services unless the customer explicitly agrees. OpenAI's API data-controls documentation lists both Chat Completions and Audio Speech as not used for training.
PrepBoss has not opted in to sharing OpenAI API data for model training.
Forvo and Merriam-Webster are lookup APIs, not model providers. Their public API terms do not contain an explicit no-training statement; the app sends them only an isolated word or short phrase, never a manuscript file, note, or voice sample. See the Forvo API information and Merriam-Webster API terms.
Current retention and deletion
| Data | Retention while a book exists | Deletion behavior |
|---|---|---|
| Original manuscript and derived data | No explicit expiry | Deleted when the user deletes the book; deleted database records remain in backups for seven days |
| Notes | No explicit expiry | Deleted when the user deletes the notes or the book; deleted database records remain in backups for seven days |
| Voice samples | No explicit expiry | Deleted when the user deletes the samples or the book |
| OpenAI API response dumps | No explicit expiry | Deleted when the user deletes the book |
| Fly.io logs (no manuscript text) | Seven days (retention documentation) | Expire under Fly.io's retention period |
| Forvo / Merriam-Webster queries | Controlled by the provider | Metadata stored in PrepBoss is deleted when the user deletes the vocabulary item or the book |
Account deletion runs the same book-deletion operation for every book owned by the account before deleting the account record.