Jev Chat

Jev Chat helps people interpret tense or ambiguous conversations and choose a reply. It estimates the other person's intent and the risk of friction before suggesting wording, though its assessments can be wrong.

• Conversation assessment: It estimates intent, emotional needs and tension, then suggests an approach such as acknowledging a feeling, apologizing or giving an explanation.
• Ranked reply drafts: It generates three responses and ranks their fit with its assessment. You can copy a draft or fill the message box; you decide what to send.
• Android chat overlay: It reads visible conversations in QQ and X, uses OCR for Feishu / Lark, and supports manual screen OCR in other apps.
• Notes and contact context: Android can use locally stored relationship profiles, notes and optional conversation history to inform replies, including context linked across apps.

The Android app is free and open source under MIT. It requires Android 11 or later on ARM64 hardware, and its interface is in Chinese. Screen OCR runs offline and screenshots stay on the phone, but analysis sends chat text and relevant context to your chosen model endpoint. It supports OpenRouter, DeepSeek, Qwen-compatible APIs and OpenAI-compatible endpoints. You supply an API key, and provider charges may apply. The developer runs no intermediary server.

Desktop editions have separate capabilities: Windows uses local OCR to read chat windows, while macOS uses a local model for intent and risk assessment and ranks replies locally. Windows requires Windows 10 1903 or later, or Windows 11; macOS requires Apple Silicon and macOS 13 or later.

Assessments focus on conflicts in close relationships. Group and work chats are less accurate, and Chinese-language judgments vary because the model was trained mainly on English.