Jev Chat (published as jev-chat/jev-chat-jarvis) is a chat decision assistant: an app that reads the conversation in front of you, works out what the other person probably means and how risky the moment is, then drafts replies you can drop into the message box with one tap. It is for people who stall over ordinary messages, a text from a manager, a landlord, a client or a date, and who want a second read before typing rather than an autopilot that answers on their behalf. The code is Kotlin under an MIT licence, the repository was created in September 2026, and it has drawn about 7,850 stars and 1,276 forks on GitHub since.
What it does
The order of operations is the whole pitch. Instead of generating a reply from the last message, Jev reads the thread first, forms a judgment about intent and risk, and only then produces reply options ranked by that judgment. You tap the one that fits, it fills your input field, and you press send yourself. The project is explicit that it never sends anything on its own.
On Android it works inside other chat apps: the materials name WhatsApp, QQ, Feishu and X. It also already runs on Windows and macOS. The project describes five apps in total across those platforms, and the platform badge lists Android, Windows, macOS and iOS.
The repository is bilingual by design, with Simplified Chinese and Vietnamese README translations alongside the English one, and an official site at chatjevs.com.
How it works
The GitHub topic list names accessibility-service and ocr, which is the usual way an Android tool reads text out of apps it does not control: it hooks the accessibility layer and reads characters off the screen. That is also why it can sit inside WhatsApp or QQ without any plugin API from those apps.
The intelligence is not in the repository. Jev is a client: you point it at a model provider, and the project's stated privacy position is that your chat text goes only to that provider and nowhere else. Nothing in the materials describes an independent audit of that claim, and the ranking of reply drafts is the project's own design rather than a measured result. No benchmark figure is published, so treat "judges risk" as what the authors say the system does, not a verified accuracy number.
Getting started
The app itself is free, but the model is not: you bring your own API key to a model provider, and that provider bills you for every conversation Jev reads and every draft it writes. The materials only mention your own provider key; they do not say whether a local model can stand in, so do not count on an offline, zero-cost setup.
Requirements beyond that are thin: the project does not list hardware requirements, memory or chip needs anywhere in the material available here.
Where the builds live:
- Android, Windows and macOS: the Releases page of the repository, github.com/jev-chat/jev-chat-jarvis, plus the download section of chatjevs.com.
- iOS: there is no ready download. You build the keyboard yourself with Xcode, which in practice means a Mac and an Apple developer setup.
Expect the Android version to ask for accessibility permission, since that is how it reads chat windows at all. The saved copy of the README is cut off before the install section, so the exact steps are the ones on the Releases page rather than anything quoted here.
When to use it / when not
The main catch is maturity, not features. WhatsApp support is still in preview, and iOS is a build-it-yourself keyboard rather than an install, so the two platforms many readers care about most are the least finished. The project is also very young: first commit in September 2026, last push in October 2026.
Use it instead of pasting screenshots into Claude or ChatGPT when the friction of copying context is what stops you, or when you want the same read applied to every thread rather than only the messages you remember to ask about.
Don't use it if you are not comfortable granting an app accessibility-level access to your private chats, if your employer forbids sending work conversation text to a third-party model, or if you want a tool that answers messages while you are away: by design, it will not.
Jev Chat is worth a serious look if you negotiate, support customers or manage relationships over chat and the cost of a badly judged reply is real. The design choice that makes it credible is the one that makes it less impressive in a demo: the human stays on the send button. Developers may also find it interesting simply as a readable Kotlin example of accessibility-service and OCR driven assistance across four platforms. Everyone else should wait on WhatsApp leaving preview, and should price the API key before installing.