01
What ISURA's AI analyzes
- The text and metadata of messages in your connected mailbox.
- Sender reputation patterns inferred from your own thread history.
- Time-sensitive language, decision cues, and risk signals in content.
- Reply patterns to surface threads waiting on you or on others.
02
What ISURA's AI never does
- Send, archive, or delete messages without your explicit click.
- Train any model on your private email content.
- Profile you, score you, or share your data with advertisers.
- Operate autonomously — there is no agent loop running on your inbox.
03
When user approval is required
Every outbound action — sending a draft, archiving, marking something as resolved, or any externally visible side effect — requires an explicit click from you. There is no "auto-send", "auto-archive", or background reply mode.
04
How operational suggestions are generated
ISURA combines deterministic heuristics (sender, recency, language cues) with LLM-based reasoning over the relevant message context. Confidence is shown alongside each suggestion. When confidence is low, ISURA says so rather than guessing assertively.
05
The line we will not cross
ISURA never takes action without user approval.