1. Frontend
Where you input the prompt

Chat with GPT-5.4, Gemini, Grok and Claude privately
Pryv routes prompts through trusted execution environments, strips PII before frontier models see it, and returns clean responses inside the same private shell.
Where you input the prompt

Your prompt is returned to The frontend
Where your request is sent, your PII is stripped and replaced with dummy data and it’s sent to State of the Art models
Your request is received by state of the art models.
Where your request is sent, your PII is replace with your real data and returned to the frontend.
Pryv inspects the prompt and redacts the categories below inside the enclave, so raw sensitive data never reaches the model provider.
Cards, bank numbers, SSNs, passports
Emails, phones, IPs, URLs, dates
Faces, OCR, EXIF, receipt markers
Scrubbing prompt text alone is not enough. Network identifiers, browser quirks, GPU surfaces, and timing signals can still correlate a user across sessions. Pryv reduces that surface by proxying model traffic and normalizing metadata inside the TEE boundary.
Totals stay while customer rows vanish from the PDF text layer.
Raster receipts run through OCR so sensitive pixels get boxed out.
CV finds faces and hits them with a 45x45 blur, leaving context intact.