Jev Chat Assistant

Overview

Jev Chat Assistant is an open-source chat copilot for Android, Windows, and macOS. It reads visible chat messages, judges intent and danger level, drafts ranked reply options, and fills the chosen reply into the input box for the user to review and send. The project emphasizes user-controlled sending, local storage for keys and notes, and bring-your-own model endpoints.

Key Features

• Reads visible chats through Android accessibility, with OCR fallback for missing text, and captures desktop chat windows on Windows and macOS.
• Judges true intent, danger level from 0 to 9, what the other person needs, whether to reply with substance, next action, subtext, and tension resolution.
• Drafts and ranks three reply options, then fills the selected option into the input box without sending automatically.
• Supports local notes and contact profiles, with pinned notes included and matched notes or recent history used only when enabled.
• Lets users configure separate judge, reply, and vision model endpoints, with presets for OpenRouter, Bocha Jev, TypeSafe, DeepSeek, and Qwen-compatible APIs.
• Open-source MIT project with Android, Windows, macOS editions and a source-only iOS keyboard.

Use Cases

• Choosing wording for sensitive or ambiguous personal messages before replying.
• Reviewing possible intent and conversational risk in one-to-one chats.
• Drafting ranked reply options in supported apps such as QQ, X/Twitter DMs, Feishu/Lark, WhatsApp, or other apps via manual OCR.
• Keeping suggested replies consistent with locally stored notes and contact context.

Getting Started

• Download the Android APK from the GitHub project page and install it on an Android 11+ arm64-v8a device.
• In Settings, add an OpenRouter API key to the judge endpoint; leave the reply and vision endpoints blank to inherit that key, and use the connection tests.
• Grant accessibility, display-over-other-apps, autostart, and unrestricted battery permissions; after an upgrade, toggle accessibility off and on again.
• Windows and macOS editions are listed on the GitHub project page with separate setup instructions.

Deployment & Requirements

• Android 11+ on arm64-v8a devices; accessibility and overlay permissions are required, and autostart plus unrestricted battery are needed on some Android skins.
• Windows 10 1903 or later, or Windows 11.
• macOS 13 or later on Apple Silicon.
• A configured model endpoint and API key for judging, drafting, and optional vision requests; provider charges may apply.
• The iOS keyboard has no App Store build and must be built from source with Xcode.

Before You Adopt

• License: MIT. Review its terms before using, modifying, or distributing the project.
• Chat text is sent to the model endpoint you configure during analysis; the author runs no server and other data stays local.
• Model calls require your own API key and may incur charges from the model provider.
• Judgments are mainly calibrated on English and close-relationship conflicts, so Chinese, group, and work chats may be less accurate.
• Some Android skins can freeze the app and make the floating bubble disappear; autostart and unrestricted battery settings help.
• Using the app inside third-party messaging apps may violate their terms and risk account restrictions or bans.