Skip to content

Latest commit

 

History

History
332 lines (284 loc) · 18.5 KB

File metadata and controls

332 lines (284 loc) · 18.5 KB
layout default
title SafeScribe — Privacy-First AI Transcription
description Your transcription app keeps your recordings for days. Ours? Zero seconds. RAM-only processing, on-device audio optimization for accuracy, deleted the moment you receive your text.
lang en
🔒

SafeScribe

Your transcription app keeps your recordings for days.
Ours? Zero seconds.

RAM-only processing. On-device audio optimization for accuracy. Deleted the moment you receive your text.

Privacy isn't a policy we follow — it's a structure we built. You can't subpoena data that was never stored.

Coming soon to App Store and Google Play.


The Difference

How SafeScribe Compares

Feature SafeScribe Otter.ai Fireflies Good Tape
RAM-only processing Yes No No No
Audio deleted immediately Yes 30 days Manual Yes
Transcript deleted from server Yes No No No
Email stored on server No (hash only) Yes Yes Yes
On-device audio optimization Yes No No No
On-device preprocessing Yes No No No
Data available for subpoena No Yes Yes Partial

How It Works

Three Simple Steps

1
🎤

Record or Pick a File

Record directly in the app or choose an audio/video file from your device. Supports MP3, WAV, M4A, MP4, and more.

2
🧠

AI Transcribes in Memory

Your audio is processed by a state-of-the-art speech recognition AI on our servers — entirely in RAM. Nothing is ever written to disk.

3
📄

Get Your Text, Data Deleted

The transcript is delivered to your device with integrity verification. All server data is immediately and permanently erased.


Our Commitments

14 Privacy Promises

{: #privacy-promises}

Every claim below is enforced in code — not just policy. Each links to the technical evidence.

Core Architecture

  • RAM-only processingYour audio exists in volatile memory only during transcription — never written to disk. learn more →
  • Cascade deletion + failsafeEach step deletes the previous step's data. If our code fails, data self-destructs via TTL. learn more →
  • Self-hosted AIYour audio never reaches OpenAI, Google, or any third-party service. We run our own infrastructure. learn more →
  • Integrity verificationEvery transcript carries a SHA-256 checksum so you know it arrived unaltered. learn more →
  • No re-downloadOnce you receive your transcript, nothing remains on our servers. There is no "download again." learn more →

What We Don't Collect or Do

  • Your email or name — never storedOnly a one-way hash of your account ID is kept. It cannot be reversed to reveal who you are. learn more →
  • Your IP address — never loggedUsed briefly for rate limiting (as a hash), then discarded. Never appears in any log. learn more →
  • Your recordings — never used for trainingYour audio is deleted immediately after transcription. No copy, no archive, no training dataset. learn more →
  • No profiling or automated decisionsWe never score, rank, or make automated decisions about you. learn more →
  • No behaviour trackingWe use zero analytics SDKs. No Mixpanel, no Firebase Analytics, no trackers of any kind. learn more →
  • No contacts, location, or biometricsWe request only microphone access — nothing else from your device. learn more →
  • Payment details — never seen by usPayments handled entirely by Apple or Google. We receive only a purchase receipt. learn more →

How We Protect What Must Move

  • AES-256 + hardware-backed keysTranscripts on your device are encrypted. Keys stored in iOS Keychain or Android Keystore. learn more →
  • Crash reports: off by defaultOptional. If enabled, personal data is automatically stripped before anything leaves your device. learn more →

Who Needs This

Built for People Who Can't Afford Leaks

📰

Journalists

Your sources are safe — not because we promise, but because we can't access them. Zero-retention means zero risk of compelled disclosure.

⚖️

Lawyers

Attorney-client privilege-grade transcription. No discoverable data on our servers — ever. Nothing to subpoena, nothing to breach.

Activists

Surveillance-proof transcription. No logs, no profiles, no traces. Your voice becomes text and the rest disappears.

💼

Professionals

Fast, accurate transcription that forgets everything. Meeting notes, interviews, dictation — with complete peace of mind.


Accuracy You Can Trust

Accurate, Even in Noise

Every audio file passes through an optimized pipeline — on your device and on our servers — to ensure the highest possible accuracy.

📱

On-Device Optimization

High-pass filtering, leading-silence trimming, LUFS loudness normalization (-16 LUFS, optimized for ASR), peak limiting, and 16 kHz resampling — encoded as lossless FLAC before upload.

🛡

Voice Activity Detection

Silero VAD identifies speech segments and eliminates AI hallucinations in silent passages by restricting transcription to detected speech regions only.

Built on established standards. Our pipeline relies on proven audio engineering standards and peer-reviewed research — not arbitrary defaults. The result: transcription accuracy you can rely on, even in challenging environments.

Pricing

Simple, Pay-As-You-Go

{: #pricing}

No subscriptions. No hidden fees. Buy credits and use them whenever you want. Unlike subscription services, you never pay for time you don't use.

$0.99
1 hour
$0.0165/min
$6.49
7 hours
$0.015/min
$12.99
15 hours
$0.014/min
$23.99
30 hours
$0.013/min
$44.99
60 hours
$0.012/min
$84.99
120 hours
$0.012/min
If our server fails to complete a transcription, you are never charged. You only pay when a transcription completes successfully on our end. Payments are handled entirely by Apple App Store or Google Play — we never see your payment details.

<script type="application/ld+json"> { "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [ {"@type": "Question", "name": "What happens to my audio after transcription?", "acceptedAnswer": {"@type": "Answer", "text": "It's gone permanently. Your audio is processed entirely in server RAM and deleted the instant your transcript is delivered. No disk copy, no backup, no recovery."}}, {"@type": "Question", "name": "Do you use my recordings to train AI?", "acceptedAnswer": {"@type": "Answer", "text": "No. Your audio is used only for generating your transcript. Never stored, analyzed, or used for model training. We run our own AI infrastructure."}}, {"@type": "Question", "name": "Which languages are supported?", "acceptedAnswer": {"@type": "Answer", "text": "The open-source Whisper model supports 99 languages with automatic detection. You can also manually select the language for improved accuracy."}}, {"@type": "Question", "name": "What file formats can I use?", "acceptedAnswer": {"@type": "Answer", "text": "MP3, WAV, FLAC, M4A, OGG, Opus, AAC, WMA, MP4, and MOV. Files up to 50 MB and 2 hours."}}, {"@type": "Question", "name": "How is my data encrypted?", "acceptedAnswer": {"@type": "Answer", "text": "TLS 1.2+ encryption in transit. AES-256 encrypted containers on your device with keys in iOS Keychain or Android Keystore."}}, {"@type": "Question", "name": "Do you log my IP address?", "acceptedAnswer": {"@type": "Answer", "text": "No. IP addresses are never stored. The logging system drops IPs before any output. We use a one-way SHA-256 hash for internal operations."}} ] } </script>

FAQ

Frequently Asked Questions

What happens to my audio after transcription?
It's gone — permanently. Your audio is processed entirely in server RAM (volatile memory) and deleted the instant your transcript is delivered. There is no disk copy, no backup, and no way to recover it. Even a physical seizure of the server would yield zero audio data.
Do you use my recordings to train AI?
Absolutely not. Your audio is used for one purpose only: generating your transcript. It is never stored, analyzed, or used for model training or improvement. We run our own AI infrastructure — your data doesn't go to any third-party AI service.
Do you see my payment or card details?
No. All payments are processed entirely by Apple App Store or Google Play. We receive only a purchase receipt for balance verification — we never see, receive, or store your card number or any payment details.
Which languages are supported?
The open-source Whisper speech recognition model supports 99 languages with automatic language detection. You can also manually select the language before transcription for improved accuracy.
What file formats can I use?
MP3, WAV, FLAC, M4A, OGG, Opus, AAC, WMA, MP4, and MOV. Files up to 50 MB and up to 2 hours long.
How is my data encrypted?
In transit: TLS 1.2+ encryption ensures your data is protected between your device and SafeScribe servers. On your device: transcripts are stored in AES-256 encrypted containers with keys protected by your phone's secure hardware (iOS Keychain / Android Keystore).
Can I delete all my data?
Yes. You can delete individual transcripts or your entire account from within the app. Account deletion permanently removes your sign-in identity, all local data, and remaining credit balance. Since the server retains no personal data after processing, there's nothing to delete server-side.
Do you log my IP address?
No. IP addresses are never stored or logged. Our architecture enforces this at the code level — the logging system is explicitly configured to drop IPs before they reach any log output. We use a pseudonymous identifier (a one-way SHA-256 hash of your sign-in identity) for all internal operations. This hash cannot be reversed to reveal your identity.
How do I verify your privacy claims?
We publish our full Security Architecture and Data Protection Impact Assessment for technical transparency. You can also use a network analysis tool (e.g. Wireshark) to confirm all traffic is TLS-encrypted — no email, name, or personal identifiers appear in any request.
How does on-device preprocessing improve accuracy?
Before upload, SafeScribe applies an on-device audio filter chain: 80 Hz high-pass filtering removes rumble and background hum, LUFS loudness normalization (-16 LUFS) optimizes audio levels for Whisper, and peak limiting prevents clipping. Combined with Voice Activity Detection on the server, this ensures you get clean, accurate text even from challenging recordings.

App Store Privacy Labels — Our privacy nutrition labels on the App Store and Google Play reflect exactly what is documented here.

Responsible Disclosure — Found a vulnerability? Contact security@safescribe.dev.

SafeScribe uses OpenAI's Whisper model weights via the open-source faster-whisper library (SYSTRAN, MIT license). SafeScribe is not affiliated with, endorsed by, or sponsored by OpenAI.