Overview & result
jev-chat shows jev-chat-jarvis, built on Jev: A reply co-pilot for phone chat apps: it reads the screen and drafts, you send.
From the original post
jev-chat-jarvis by @jevchat: A reply co-pilot for phone chat apps: it reads the screen and drafts, you send.
How Jev fits in the loop
- An adapter per chat app turns the current window into a title and a message list; screen capture and offline OCR are the fallback.
- Jev answers all choice, score and yes/no questions in one request, in about a second by the author's figure.
- A generative model drafts three candidate replies and Jev ranks them.
- The chosen reply is filled into the input box; the app never sends it.
How to reproduce
- Review the source documentation and repository at https://github.com/jev-chat/jev-chat-jarvis.
- Configure environment with TypeSafe Jev / System One API keys.
- Run test prompts against the decision endpoints.
Why this build matters
Shows Jev in practical use for developer tools and apps.
Limitations
- Group chats are analysed as one-to-one, so the settings do not fit them.
- The author notes Jev is trained mainly on English and suggests calibrating on your own labelled chats.
- Knowledge-base retrieval is tag and title matching, not semantic search; some Android ROMs may kill the background service.