ReplyMint

A privacy-first AI reply assistant that writes the perfect reply directly inside any app — using your context, voice, and business memory.

ReplyMint is an Android-first floating bubble. You tap it inside WhatsApp, Gmail, LinkedIn, Instagram — any app with a text box — and it drafts one best reply directly into the input field. You review and tap Send yourself. ReplyMint never sends anything.

Two modes

	Personal	Professional
For	Everyday quick replies	Founders, sales, consultants, recruiters
Does	Fix, Voice, simple reply from screen	Full Reply Brain: business rules, voice, contact memory
Memory	None (stateless)	Private, editable, deletable Reply Brain
Price	Free / $4.99/mo	$19–$49/mo

Paying (either mode) also unlocks cloud transcription (Deepgram, pro-gated server-side); free accounts dictate with the on-device engine — audio never leaves the phone.

The three actions

1. Auto Reply Draft — reads the visible conversation, writes one best reply.
2. Voice — you speak an instruction ("tell him I'll be late but stay friendly"), it rewrites.
3. Fix — cleans up grammar/clarity of what you already typed.

How it works (Android)

Two Android capabilities do the heavy lifting:

• SYSTEM_ALERT_WINDOW draws the floating bubble on top of every app.
• AccessibilityService reads the visible conversation on tap only, and writes the finished draft into the focused input field.

Tap bubble → Accessibility reads screen → pick action → backend → LLM → one draft
          → draft written into the input box → you review → you tap Send

Platform scope

• Android — the primary product. Full experience (bubble + all three actions + Reply Brain).
• iOS — a separate, later project: a Grammarly-style keyboard doing Fix + Voice only. A keyboard cannot read the incoming message, so Auto Reply Draft is Android-only. Not in this repo yet.

Repo layout

replymint/
├── docs/            Architecture, roadmap, privacy, Reply Brain spec, dev setup
├── android/         Native Kotlin app (the product)
├── desktop/         Tauri menu-bar app (Mac first): hotkey dictation into any app
└── backend/         Lightweight Hono + Claude API server

Getting started

• Android: see docs/DEV_SETUP.md
• Backend: see backend/README.md

Privacy promise (enforced in code, not just marketing)

• Reads the screen only when you tap the bubble — never a background listener.
• No send capability at all — the app can only write into the box.
• Personal mode is stateless. Professional mode stores derived memory, not raw chats.
• Reply Brain is fully editable and deletable.
• No ads, no data selling, no training public models on your data.

See docs/PRIVACY.md for how each promise is implemented.