Megadose AI progress, ranked and analyzed.

Finderchangchang/jev-chat-jarvis

· GitHub · LLM repos ·
An Android assistant reads visible chat screens via accessibility, rates intent and risk, and drafts three replies without sending them.

The repo says it has full-chain tests on WeChat, QQ, and X/Twitter DMs, with partial Feishu/Lark capture already wired in. Its design uses one shared core and small per-app adapters that turn the current chat window into a title plus recent messages. It says replies are only filled into the input box, never auto-sent, and it does not handle transfers, red packets, or payments. The author notes limits around Chinese ROM background killing, Feishu’s custom-rendered message body, X being verified only in a Chinese UI, and WeChat accessibility obfuscation possibly breaking later. GitHub · LLM repos' note

score 4

Categories: OSS & Tools