How it works
Built around you, not around engagement.
Most apps are designed to keep you coming back. Streaks guilt you into opening the app. Notifications interrupt you at the wrong moment. Badges reward frequency over depth. Eir doesn't do any of that.
The only metric that matters here is whether you understand yourself a little better today than yesterday. The design follows from that. Everything else — the encryption, the AI, the pattern analysis — is infrastructure in service of one thing: a space that's genuinely yours.
Three things Eir does well
End-to-end encrypted
Your entries are encrypted before they leave your device. No one — including us — can read them.
AI that reflects, not prescribes
Eir asks questions that help you hear yourself more clearly. A mirror, not a therapist.
Patterns over time
Recurring themes, energy rhythms, relational patterns — surfaced slowly, without algorithms shouting at you.
End-to-end encryption
Every entry is encrypted in your browser before any network call. The algorithm is AES-256-GCM — the same standard used by banks and password managers. The encryption key is derived from your credentials using PBKDF2, which means it never exists anywhere except in your active session.
What reaches our server is ciphertext — a string of characters that is mathematically meaningless without your key. We store it. We can't read it. If our database were ever compromised, your entries would still be unreadable.
This isn't a privacy policy promise. It's an architectural fact.
AI that reflects
The AI is not a therapist. It doesn't diagnose, prescribe, or tell you what to do. It doesn't say "I'm so sorry you're feeling that way." It asks the kind of questions a thoughtful person would ask — the ones that open doors you didn't know were there.
Responses stream in real time. You see words appear as they're generated, not after a pause. This matters more than it sounds — a response that arrives gradually feels like a conversation, not a lookup.
The AI's only job is to help you hear yourself more clearly. It's a mirror. It reflects what you wrote back in a way that might reveal something you already knew but hadn't said out loud.
Patterns over time
After enough entries, recurring shapes start to emerge. Themes you return to. Times of day when your writing is heavier. Relationships that appear and disappear. Eir surfaces these slowly — not as an algorithmic report, but as quiet observations.
Pattern analysis runs on-demand, on content that's been decrypted locally. Nothing leaves your device as raw text for analysis. Only lightweight derived signals are stored — "high negative sentiment on this date", not the words themselves.
The framing matters: "Something keeps showing up in your writing" — not "Our system detected."
The full flow
Write
You open the editor and write — as much or as little as you want. No formatting, no audience.
Encrypt
Before anything leaves your device, your entry is encrypted in the browser using AES-256-GCM. The key is derived from your credentials — never sent anywhere.
Store
Only the ciphertext reaches Supabase. The server holds an unreadable blob. No plaintext, no moderation layer, no training pipeline.
Read
When you open an entry, the ciphertext is fetched and decrypted locally in your browser. It never travels anywhere as readable text.
Reflect
If you ask for a reflection, the decrypted content is sent to the AI only for that request — streamed back to you in real time, never cached server-side.
Surface patterns
Over time, lightweight signals are extracted client-side: themes, timing, sentiment. These are stored as encrypted metadata — never as raw content.