Made for Swiss self-employed & SMEs
A tense client message lands on WhatsApp. You read it twice and still can’t tell: are they annoyed, or just in a hurry? Jev, a free and open-source app, is built for exactly that moment. It reads the conversation, works out the mood and the risk, and then offers you ready-made replies — ranked by that judgment. You still press send yourself.
What Jev actually does
Most AI writing helpers work the other way round: you ask for a text, the machine produces one. Jev starts a step earlier, with a judgment. Before any draft exists, it reads the visible conversation and asks three quiet questions: What does this person really mean? How risky is this moment? What would be a good way to answer?
Only then does it write. You get several reply drafts, ranked and matched to that reading. Tap one and it drops into your message box, ready to edit. Jev never sends anything on its own.
On Android it runs as an overlay inside WhatsApp (plus X, QQ and Feishu). Separate Windows and macOS apps read the chat window with screenshots and on-device text recognition.
The honest catches
Jev is young software, and it asks for wide access. Before installing, know these points:
- WhatsApp support is a v0.1.0 preview — early, and it shows.
- There is no iPhone app in the App Store; on iOS you would have to compile the keyboard yourself.
- To read a conversation, the app uses strong permissions: the Android accessibility service, or screenshots plus text recognition on the desktop.
- Your chat text goes to whichever model provider you pick, with your own API key. The project runs no server of its own — but the words still leave your device.
- The app is free under an MIT licence; each judgment costs whatever your model provider charges per request.
The claims about reach and quality come from the project itself; no independent test exists yet.
What this means for client confidentiality
If you are self-employed in Switzerland, your WhatsApp threads are often half a client file: names, addresses, deadlines, prices, sometimes health or legal details. The revised Swiss data protection law does not forbid AI helpers. It simply holds you responsible for what leaves your devices, and for choosing tools and providers with care.
So the practical question is not “is Jev allowed?” but “which conversations may it see, and where do they travel?” Treat the model provider like any subcontractor: which company is it, where does it process the text, and may its terms use your input for training?
One design decision deserves credit: Jev drafts, you send. A draft sitting in a text box is a suggestion, not a disclosure. The sensitive moment happens earlier, when the app reads the conversation. That is the step to think about — not the send button.
A sane way to try it
- Get the Android, Windows or macOS app from the project’s GitHub releases page — the code is open, so anyone can read it.
- Use an API key from a model provider you have already checked, or one you already use elsewhere.
- Test it on a conversation that carries no client data: your own second phone, a family group, a colleague.
- Keep one rule: Jev proposes, you decide. If a thread contains anything you would not forward to a stranger, answer it without the helper.
For tense client moments, a second reading of the room is genuinely useful — just keep the choice of what leaves your phone in your own hands.
