Security & Data Privacy

Your content. Your control.

We understand that your books are your most valuable assets. Here is exactly how we protect them, in plain language, not legal boilerplate.

Our Core Guarantees

Isolated per-publisher namespace

Your book content is stored in a vector namespace exclusive to your account (pub_{your_id}_book_{book_id}). No other publisher, no QuantumRead employee, and no system process can query your namespace.

Content is never reproduced

We convert your text into mathematical vector embeddings, numerical representations that cannot be reverse-engineered back to the original words. The AI answers questions about your content; it cannot output paragraphs or chapters from your book.

Zero model training on your data

The AI model we use is a third-party API (OpenRouter). Your book content is sent as temporary context for each question, then discarded. It is never used to fine-tune, train, or update any model, ours or anyone else's.

Delete on demand, instantly

Click Delete on any book in your dashboard and its vector data is removed immediately. No 30-day grace period, no backup retention window. We also honor deletion requests via email within 24 hours.

Encrypted in transit and at rest

All data is transmitted over TLS 1.3. Vector store data is encrypted at rest using AES-256. API keys are hashed using SHA-256 and never stored in plaintext, not even we can recover your key.

API key access control

Each widget request must present a valid, active API key tied to your account. You can revoke a key instantly. Rotating keys takes under 30 seconds and does not require any code changes.

How Book Ingestion Works

When you upload a PDF or ePub, here is exactly what happens, step by step:

  1. 1
    Upload, Your file is uploaded over TLS to our processing server via a short-lived presigned URL.
  2. 2
    Extraction, Text is extracted from the file. Images and formatting are discarded, only the semantic text content is retained.
  3. 3
    Chunking, The text is split into ~300-word overlapping chunks. Your book typically becomes 150–400 chunks depending on length.
  4. 4
    Embedding, Each chunk is converted into a 1536-dimensional vector (a list of numbers). The original text of each chunk is stored alongside its vector for retrieval.
  5. 5
    Storage, Chunks and vectors are stored in your private ChromaDB collection. The original uploaded file is deleted from our servers after processing.
  6. 6
    Query time, When a reader asks a question, the 3–5 most relevant chunks are retrieved and sent to the AI as context. Only those chunks, not the whole book, are used per question.

How This Compares to Other Platforms

Publishers already upload full book content to distribution platforms. QuantumRead is not a new risk, it is a more controlled one.

PlatformHolds full contentIsolated to youDelete on demand
Amazon KDPShared infrastructureSlow process
Apple BooksShared infrastructureSlow process
Google BooksShared infrastructureSlow process
Scribd / OverdriveShared infrastructureSlow process
QuantumRead

Data Processing Agreement

We are happy to sign a Data Processing Agreement (DPA) before you upload any content. This is standard practice for enterprise publishers and provides legal binding for all the guarantees on this page.

Our DPA covers: data isolation, retention limits, sub-processor disclosure, breach notification timelines, and your right to audit. Most publishers receive a signed DPA within 48 hours of request.

Request a DPA

Questions about security? security@quantumread.space

Ready to start? Open Publisher Portal