In Trust6, a message never leaves as-is. The flow is always the same: write, analyze, review, send. Here is a first conversation, step by step.
1. Write
Open your vault, create a conversation and write your message normally, with real names and real figures:
Jean Tremblay from Métal ABC inc. offers $25,000.50 for the project in Montréal.
Notice: the Send button is inactive and the indicator says the text has not been analyzed. The app refuses to send anything it has not analyzed. That is not a warning, it is a guard at the core of the system.
2. Analyze
Press Enter (or click Analyze). The analysis panel opens to the right of the chat:
- The detection count by type appears at the top.
- Each detection is highlighted in your text and represented by a dismissible chip.
- The pseudonymized preview shows the exact text that would be submitted to the model, tokens included (
PERSON_001,ORG_001,AMOUNT_001).
The very first analysis after launching the app takes about fifteen seconds: the local engine loads into memory, and the indicator says so. Subsequent analyses take a second or two.
3. Review
You decide. For each detection:
- The × on a chip dismisses the detection (the term would then leave unmasked, an assumed and logged choice).
- A missed term is caught by selecting it, right-clicking, and choosing its type: it enters the vault dictionary and will be masked for life, automatically.
If you edit your text after analyzing, the send button greys out and the app asks you to re-analyze: the approved text and the sent text cannot diverge.
Full details in Review detections before sending.
4. Send
Click Send. The answer streams in, and Trust6 restores the real values on screen. Under your message, the “Pseudonymized before sending” badge confirms how many entities were masked, and you can inspect word for word what actually went over the network.