Skip to content
Jev Chat Assistant

Practical guide

Set up Jev Chat Assistant: choose a platform and generate your first reply options

A practical Android setup guide covering model endpoints, accessibility, overlay permissions, costs and where chat data goes. Desktop editions have separate instructions.

Finderchangchang ·

Choose the platform before following setup instructions

The main Jev repository links to separately maintained Android, Windows and macOS editions. The accessibility, overlay and OCR steps below describe Android. Follow each desktop project’s own instructions for Windows or macOS.

PlatformRequirements currently listedNext step
AndroidAndroid 11+, ARM64Follow the steps below
WindowsWindows 10 1903+ / 11Choose the Windows project from the main repository
macOSmacOS 13+, Apple SiliconChoose the macOS project from the main repository

An iOS edition and a browser version are not currently available. Start from the GitHub project homepage and check the release notes for your platform.

1. Get and install the Android package

Use the Android entry in the main repository. Check the Android version and CPU architecture first; this guide covers ARM64 devices.

A previous debug build may have a different signature and prevent an in-place update. Uninstalling removes settings and keys, so account for anything you need to retain first. After a normal upgrade, turn accessibility off and on again to rebind the service if necessary.

2. Configure and test the model endpoints

In the app’s endpoint settings, there are separate cards for judgment, reply generation and vision. The current README describes starting with your OpenRouter API key in the judgment card and leaving the other two blank to inherit it. You can also configure each route independently.

  1. Make sure the provider address, model name and API key match.
  2. Run the connection check on the endpoint card.
  3. Once the endpoint works, test screen reading in a supported chat.

Free open-source software does not imply free model calls. Your chosen provider sets API charges. Keep keys out of screenshots, issues and public documents.

3. Follow the app’s permission guide

Reinstalling may reset overlay permission. Permissions allow capture and display; they do not replace model endpoint configuration.

4. Check the full flow with non-sensitive content

Try a supported chat with non-sensitive text. Compare the recognized words with the screen, inspect the assessment and reply options, then copy or fill a candidate. Review it before pressing Send yourself.

The website records verification for QQ and X direct messages, and OCR-assisted text capture for Feishu / Lark. This does not establish support for every app version or interface language. Read the compatibility notes first.

Troubleshoot the stage that fails

Know where the data goes

Screenshot recognition happens locally on Android. When you request analysis, chat text and enabled background context are sent to the model endpoint you configure. Local OCR does not make the whole process offline. Review the provider’s policy and the Jev privacy summary.

This guide is based on public project documentation, not an independent test across all devices. Sources: the project README and website compatibility notes.