Repo

jev-chat-jarvis

Provides reply suggestions inside chat apps by reading on-screen messages, scoring intent/risk, and inserting a chosen draft without sending it.

GitHub ↗Sep 21, 2026

  • llm
  • android
  • accessibility
  • ocr
Pipeline diagram showing OCR-to-reply workflow with local privacy design
Pipeline diagram showing OCR-to-reply workflow with local privacy design

What it is

Android chat copilot that reads visible conversations in other apps, judges intent/risk, and drafts replies without hooking, app mods, or auto-send. Adds an on-device chat copilot to QQ, X, Feishu, and other apps: it reads visible messages via accessibility/OCR, analyzes intent and risk, and fills ranked reply drafts into the input box for manual send.

Key results

  • Judgment returns intent, danger 1-9, action, and reply urgency in ~1 s
  • Supports QQ and X end-to-end; Feishu uses offline ML Kit OCR fallback
  • Never auto-sends; fills via ACTION_SET_TEXT, falls back to clipboard paste

Shows a cross-app chat assistant can be built from accessibility plus local OCR, without invasive app integration.

One useful ML email a week.

Worth your time, not all of your time.

We track new ML work across papers, code, and data, then keep only the pieces worth a closer look. Selected, not collected.

Papers.Data.Code

The Weekly ML Signal

Selected, not collected.

One useful ML email a week: the papers, repos and datasets worth your attention.

Worth your time, not all of your time.