Privacy & Security Architecture.
CandorVault is engineered from the operating system kernel up to be 100% invisible, leak-free, and ephemeral. We never train models on your private conversations.
Hardware Screen Exclusion
Using the Windows Desktop Window Manager (DWM) exclusion primitives (`WDA_EXCLUDEFROMCAPTURE`), your teleprompter overlay is physically omitted from screen captures, Zoom, Teams, and Google Meet video pipelines.
Zero-Retention RAM Buffer
Audio frames and conversation transcripts exist exclusively in volatile memory buffers. When a session ends or the double-escape emergency key is pressed, memory blocks are immediately zero-overwritten.
Direct Digital Audio Loopback
No virtual sound drivers or suspicious third-party virtual audio cables that trigger proctoring alerts. Audio is captured via direct WASAPI loopback without altering your microphone chain.
1. What Data We Process
CandorVault processes speech input strictly to synthesize real-time conversational hints, technical code suggestions, and interview answers. Incoming audio streams are transcribed in sub-300ms chunks and sent directly to stateless LLM inference nodes.
- No raw audio recordings are stored on our servers.
- Transcripts are stored in your encrypted local session cache only if Ephemeral Mode is disabled.
- You can toggle "Ephemeral Zero-Retention Mode" at any time from your session settings.
2. Emergency Privacy Mechanisms
In high-stakes interviews, unexpected situations happen. CandorVault includes an emergency instant hide trigger: pressing Escape twice immediately hides the floating HUD, zeroes active audio buffers, and pauses loopback listening.
3. Payment Security & PCI-DSS
All payment transactions are handled through Safaricom Daraja 2.0 (M-Pesa STK push) and Paystack (PCI-DSS Level 1 certified payment gateway). CandorVault never stores credit card details or bank account credentials on our infrastructure.