Privacy and security
Your data stays yours
dotosi sits between your people and the AI. Here is exactly what each side sees, and what we promise.
- If protection is down, nothing goes out
- Never used to train AI
- Encrypted with AES-256
One message, as each side sees it
Our promises
Built in, not bolted on
The answers to the three questions every company asks before it lets staff use AI.
What reaches the AI?
The AI gets placeholders, not real details
Names, numbers and other sensitive details become placeholders before a message goes to any AI. Your employee checks what was covered first, and anything they choose to send uncovered is logged.
No outside company reads your data
The part of dotosi that finds sensitive data runs on private systems. No outside service ever sees the uncovered text.
If protection is down, nothing goes out
If the protection layer is ever unavailable, the message simply stops. Nothing is sent and nothing is stored.
What is kept, and for how long?
Real details are locked away
The real values are stored encrypted, so even someone holding the database cannot read them.
No one can build a profile from your chats
The same person gets a different placeholder in every chat, so nothing can be pieced together across conversations.
Delete it, and it's gone for good
Erase a company's locked values and the real details behind every placeholder are gone forever. The masked record stays for your own compliance.
Who stays in control?
Your AI accounts stay yours
Each company uses its own AI provider accounts. Your messages never run on an account shared with anyone else.
Every exception is on record
If someone chooses to send a value uncovered, or views a real value, it is logged and admins can be alerted.
We never train AI on your content
dotosi never trains any model on your content, and the AI provider receives masked text unless your employee chooses otherwise.
For your IT and security team
The technical detail
Every promise above maps to the code that implements it. A full security-architecture document is available on request, so your reviewers' due diligence starts from specifics.
- Encryption
- Real values encrypted with AES-256-GCM in the application; the database holds ciphertext only. Keys are versioned and rotatable.
- Detection
- Fixed patterns (IDs, IBANs with checksum validation, cards) plus an AI model, Arabic and English, on private infrastructure with no public access.
- Fail-closed
- If detection is unavailable the request stops: nothing is sent to a model and nothing is stored.
- Raw text
- Held in memory while the request runs and never stored. What is saved is exactly what was sent: masked text, plus any value the employee deliberately un-masked on the review screen, which is logged.
- Placeholders
- Scoped per conversation, and each value is bound to its conversation.
- Provider keys
- Brought by each company, encrypted and scoped to that company. Never a shared key.
- Audit
- Masked record of every conversation; every unmasked send and every reveal is logged, with admin alerts.
- Self-hosted
- Keys created on your server; no telemetry, no phone-home.
Need data to stay in-house?
Run dotosi in your own cloud or on-premise, and only masked text ever goes out.
Let your whole company use AI, safely.
See dotosi on your own kind of data in a 20-minute demo. No commitment.
Request a demo and our sales team will contact you to schedule it.