À propos de Selyf

Selyf est un assistant e-mail IA pour Gmail conçu autour de la sécurité. Il existe pour répondre à une seule question — qu'est-ce qui, dans cette boîte de réception, nécessite réellement mon attention — sans demander plus d'accès que cette question ne l'exige.

Il s'agit aujourd'hui d'un logiciel en phase précoce (version 0.1.0) : la synchronisation en lecture seule, l'analyse IA et le flux de rédaction de brouillons sont réels et fonctionnels. L'envoi direct de courrier depuis Selyf n'est pas encore implémenté — voir le changelog pour ce qui a réellement été livré.

  • Read-only, to start

    The current version can only read your inbox. It cannot send, delete, or change anything — not because we haven’t gotten to it, but because giving that access is a deliberate, later decision.

  • Minimal data, on purpose

    We keep only what the product needs to function, for only as long as it needs it, and full message bodies are never stored — only metadata, and only what’s needed to show your inbox.

  • You approve every send

    AI drafts are suggestions. Nothing ever reaches your contacts without you copying it, pasting it, and sending it yourself — there is no auto-send path today.

Ce qui distingue Selyf

  • Read-only Gmail access

    gmail.readonly is the only OAuth scope requested. There is no code path anywhere in the product that can send, delete, or modify a Gmail message.

  • AI-based email analysis

    Each thread you open can be analyzed on demand, instead of you reading every subject line to figure out what matters.

  • Priority classification

    Urgent, Important, Normal, or Low — a quick read on how much attention a thread actually deserves right now.

  • Confidence shown, not hidden

    Every AI result ships with a confidence score next to it — never rounded up to look more certain than it is.

  • Draft generation, not auto-send

    An editable reply draft you can regenerate in a different tone — copy it to your clipboard yourself, nothing is sent automatically.

  • Built against prompt injection

    Email content is placed inside an escaped, delimited data block in the AI request — never treated as an instruction, by construction, not just by design intention.