Categorization

How Strind decides what should reach you now, later, or not at all.

Today, Strind works from four simple questions: where the message came from, who sent it, what it says, and when that person is allowed to reach you. It uses a trust-first system built from explicit message signals, lightweight private heuristics, and a small amount of self-hosted supporting intelligence on Strind servers. Message content is encrypted at rest and is not sent to third-party AI providers for triage.

What happens first

Step 1

Strind checks whether the sender is allowed to reach you at all. Unknown senders go to quarantine first instead of landing on the device.

Step 2

For allowed senders, it reads the explicit signal in the message like !!, !, or no prefix.

Step 3

It applies a lightweight smart pass for obvious urgency, timing, and repeated attempts before assigning the final bucket.

The 4 Ws behind the decision

  • Where: which channel was used, such as SMS, Telegram, Email, or WhatsApp.
  • Who: whether the sender is trusted, unknown, promoted, or hidden.
  • What: the urgency markers and timing language inside the message itself.
  • When: whether that person should be allowed to break through at that time.

Trust comes before priority

  • Trusted or explicitly allowed senders can reach the live device queue.
  • Unknown senders are quarantined first so you can review them without being interrupted.
  • Quarantined messages are held separately until you trust the sender, dismiss them, or they expire.

What the smart pass looks for

  • Clear urgency wording such as urgent, asap, emergency, or similar direct language.
  • Time-sensitive wording such as today, tonight, in 10 minutes, or a near-term time.
  • Repeat activity from the same sender over a short window, which can be a signal that something genuinely needs attention.
  • Common automated patterns such as receipts, newsletters, verification codes, or order confirmations so those messages do not get accidental priority boosts.

About local AI

Strind does use a small self-hosted local model on Strind servers for supportive features like quick reply suggestions. That is separate from the core priority triage path. The actual urgent / important / later decision remains grounded in sender trust, explicit message signals, and private heuristics that are easier to explain and audit.

Privacy in the routing path

Strind has to read message content briefly to understand urgency, quarantine unknown senders, and run safety checks. That processing happens inside Strind systems, then message content is stored encrypted. Routine business-console views show metadata and reasons instead of turning private messages into an operator inbox.

What it does not do

  • It does not send message bodies to OpenAI or any other third-party model API for categorization.
  • It does not silently invent priorities out of nowhere. Trust, sender signal, and a small set of private rules still matter most.
  • It does not override your deny rules. If you blocked a sender, that message stays out.

Why Strind works this way today

This version is intentionally conservative. It gives you something meaningfully smarter than a plain inbox while keeping the system explainable, private, and easy to audit. Every decision is recorded with reasons so Strind can tell you why a message landed where it did.

Short version: trust comes first, message signal comes second, lightweight smart rules come third.

Where this can go later

The long-term goal is stronger categorization without weakening user trust. That could mean richer self-hosted intelligence on Strind servers, or more private on-device categorization as hardware improves. For now, the system favors privacy, predictability, quarantine-first filtering, and clear behavior over heavier AI.