Last updated: November 7, 2026
SolScribe transcribes uploaded audio on cloud GPU workers and attempts to delete staged audio when processing ends. Failed deletes are retried later. Transcripts and derived records stay in your account, encrypted at rest, until you delete them. AI features can send text and related context to a language-model provider, and provider copies follow their own retention rules.
This policy covers the hosted service at app.solscribe.ai and the marketing site at solscribe.ai. SolScribe's self-hosted edition runs on your own hardware, and The Self-Hosted Edition below says what it shares. SolScribe Capture for iPhone covers the iPhone app.
Do not send Protected Health Information (PHI) to the hosted service or through cloud transcription. The HIPAA tier is on hold. No vendor that processes hosted data has signed a BAA with SolScribe.
When a cloud transcription succeeds, SolScribe generates a downloadable Certificate of Destruction, but only after it confirms the staged audio was deleted or that none was staged. The certificate lists the job IDs, audio duration, estimated file size, processing tier and a SHA-256 hash of the job metadata rather than of the audio. It states where the staged copy was deleted from and the UTC time that was recorded, or that no copy was staged.
A certificate is a staging-only attestation. It does not cover the GPU worker's disk or memory, RunPod job records or any other provider copies. Failed and cancelled jobs get none. If the delete at completion can't be confirmed, the fallback sweep can issue the certificate after its own confirmed delete, stamped with the sweep's time. A recording uploaded in pieces and merged gives one download holding each piece's certificate, and it names any piece without one. Audio destruction is also written to the immutable audit log with actor, action, record ID and timestamp.
| Data | Why | Where | Retention |
|---|---|---|---|
| Staged audio | Transcription | encrypted AWS S3 for uploads over about 5 MB | Deletion is attempted when processing ends. The fallback sweep retries failed deletes. |
| RunPod job inputs and results | Cloud processing | RunPod. Inline inputs contain audio. Results can contain transcript text, timings and voiceprints. | Provider job lifetime and retrieval windows described below |
| Transcripts and derived records | Transcript review and AI features | Neon Postgres, on storage encrypted at rest (AES-256) | Until deletion; includes speaker labels, voiceprints, summaries, reports, chat history, notes, highlights and Brain entries, subject to the deletion details below |
| Account data (email, password hash) | Authentication | Neon Postgres + Neon Auth | Life of account |
| Compute ledger (minutes used) | Bill the correct number of minutes | Database | Life of account; no audio content or filenames logged |
| License purchase records | Issue and deliver your purchased license | SolScribe account database and our self-hosted Keygen instance; no audio or transcripts | Local purchase records are removed with your account. Issued license records remain in the license service to support your purchase. |
| Payment data | Process payments | Stripe for current purchases and Gumroad for supported legacy purchases; we never see card numbers | Under each payment provider's policies |
| AI requests and responses | Titles, analysis and other AI features | Anthropic by default, or the provider selected with your own key | Provider terms and account settings govern retention, with possible abuse-monitoring and legal exceptions |
| Security audit log | Record access and security actions | Account database; includes actor, action, record ID, timestamp, IP address and browser type | Each entry stays six years, including after account deletion |
| Pending staged-upload cleanup record | Remove a late staged copy | Account, job and storage identifiers in the database | While the storage outcome is unresolved, including after account deletion. The six-year audit expiry applies once cleanup is confirmed. |
| Public HIPAA waitlist | Notify you when the tier opens | Account database; email and any organization or notes supplied | 12 months from signup or sooner on request; signed-in waitlist requests instead remain in the security audit log |
| Database restore history | Recover from service failures | Neon | The configured project history window; it ages out separately from deletion in the active database |
| Operational logs and error reports | Reliability and debugging | Fly.io and Sentry | Retention varies with provider settings |
RunPod documents a default 24-hour job lifetime from submission and a separate 30-minute window to retrieve asynchronous results after completion. SolScribe does not purge RunPod's job records. These documented defaults and retrieval windows do not establish when every provider copy is erased.
The exact current account retention windows are not confirmed for Neon restore history, Fly.io logs and Sentry reports. Deleting active records does not immediately erase provider restore history or operational records. AI providers also retain requests under their own terms and account settings. We do not guarantee a deletion time for copies outside our control.
Deleting a transcript removes its associated records, including notes and highlights. Brain's separate person and project names are kept by design. A complete historical orphan purge of notes and highlights left by older transcript deletions has not been verified. Account deletion removes remaining rows tied to your account. Security audit entries and external provider records follow their separate retention rules.
Operational logs contain route patterns, status, timing, internal IDs and error text. Sentry reports remove request bodies, full URLs, cookies, IP addresses and browser details before sending. These logs and reports are configured to exclude audio and transcript text. Pushover alerts contain a fixed title and the Sentry report ID.
All data is transmitted over TLS. Transcripts are stored in Neon Postgres, whose storage is encrypted at rest with AES-256. AWS S3 encrypts staged audio at rest.
SolScribe does not use your audio or transcripts to train AI models. Processing includes transcription and the AI features below. External providers handle requests under their own terms, including the agreement for any provider account you configure.
Hosted transcription can automatically generate titles and analysis reports. Relevance scores, summaries, quality checks and Brain knowledge extraction can also run under their applicable settings and quality gates. Chat and on-demand summaries run when you request them.
The hosted default is Anthropic on SolScribe's account, with no fallback to OpenAI. You can choose OpenAI or Anthropic and save your own key through the SolScribe API; the web app has no AI settings page. With no saved key, the hosted service uses SolScribe's Anthropic account. A request that fails on your key is not retried on ours. Supported maintenance, including quality checks and knowledge backfills, resolves each account's configured provider and applies the same key and routing rules. Hosted processing does not accept local model servers.
AI providers receive transcript text or excerpts and related evidence, not audio. Depending on the feature, this includes chat messages and conversation history, relevant Brain entries from other recordings, the display name set for Brain, titles, summaries, project context and open action items. This evidence can include names and other personal information spoken in a recording. Provider retention is separate from deleting a transcript in SolScribe.
Hosted AI analysis does not currently debit transcription minutes. A provider you configure with your own key may bill you separately. The terms explain licenses, transcription credits and goodwill credits.
We rely on the following service providers to operate SolScribe:
The marketing site uses self-hosted Plausible analytics (no cookies, no personal data). The hosted application itself contains no analytics or advertising trackers.
Local transcription and a local language model keep audio and transcript processing on your hardware or local network. Selecting a cloud AI provider sends text and related evidence to that provider. Webhooks send transcripts to the configured destination. Cloud transcription sends audio to the configured RunPod endpoint and uses its configured staging bucket. License purchases send purchase details to our license service, without audio or transcripts.
Model downloads and update checks contact Hugging Face, and speaker alignment can fetch models from PyTorch or Hugging Face. These services see the server's IP address and model requests. A configured Hugging Face token identifies your account. Source installations and older containers can also contact GitHub to download a sentence tokenizer; current containers bundle it.
Current container targets disable both pyannote and ONNX Runtime usage reporting with PYANNOTE_METRICS_ENABLED=false and ORT_DISABLE_TELEMETRY=1. Source and native installations differ with library versions and settings, as do older images. When enabled, pyannote reports audio duration, speaker settings, library version and a random session ID to pyannoteAI. ONNX Runtime versions with reporting send usage events to Microsoft. These reports expose the server's IP address but contain no audio or transcript text.
Self-hosted AI uses the selected provider, with local Ollama as the Docker default and support for LM Studio. An optional OpenAI fallback is off by default. Enabling it can send text to OpenAI after an eligible failure, including a local-provider failure. Local-only mode (PHI_LLM_LOCAL_ONLY=true) blocks direct external AI requests and disables that fallback. A local gateway that forwards requests still needs its own routing controls.
Your server retains audio unless you delete it or enable auto-cleanup. Self-hosted backups can retain deleted audio and include the encryption key. Your server and backup settings determine retention.
SolScribe Capture is an iPhone app from DST Digital LLC, available on the US App Store. It analyzes microphone audio on the phone and keeps only the pieces that contain speech.
The app uploads those pieces to one destination, never both:
The app keeps these on the phone:
Questions about the app go to privacy@solscribe.ai, as listed under Legal Entity & Contact below.
We're not offering HIPAA processing yet. The managed HIPAA tier is on hold, and DST Digital LLC is not signing Business Associate Agreements (BAAs) with customers while it is. Do not send Protected Health Information (PHI) to the SolScribe cloud service. You can join the waitlist to hear if that changes.
SolScribe is operated by DST Digital LLC. Privacy questions or data requests: privacy@solscribe.ai / compliance@solscribe.ai.
We will update this policy as the service evolves and bump the "Last updated" date above.