Jev Chat Jarvis for macOS
macOS chat overlay that judges incoming-message intent and reply risk, then ranks suggested replies.
Added to Jevfast
How it uses Jev
When a supported chat app shows a confirmed incoming message, Jev classifies its intent and risk in one request. A separate reply model generates suggested replies in parallel. Jev can then rank those candidate replies in another Choice request. The user reviews the floating panel and may copy or fill a suggestion; sending remains manual. If the Jev request fails, the app switches to its local decider model.
What Jev decides
Incoming message intent and riskExample answers · not a recorded Jev response · Source ↗
Question 1 · intent
YOUR APP
INSTRUCTION
这句话的真实意图是什么？
STATE
The current message, optionally preceded by chat context. The request asks Jev to classify intent and assess the risk of replying directly.
JEV · CHOICE
- 派活
- 催进度
- 问进度
- 批评
- 要解释
- 闲聊
- 约会议
- 夸奖
Question 2 · risk
YOUR APP
INSTRUCTION
如果直接回复这句话，风险有多大？
STATE
The current message, optionally preceded by chat context. The request asks Jev to classify intent and assess the risk of replying directly.
JEV · SCORE
0
App workflow
- Read a new incoming message
The app watches the foreground chat window and starts analysis for a confirmed incoming message, subject to a two second minimum gap between analyses.
- Classify and generate suggestions
Jev judges intent and risk while a separate configured model generates reply candidates. Jev may rank the candidate replies in a second request.
- Review before replying
The floating panel displays the judgment and candidate replies. The user can copy or fill a candidate, but the app does not send messages.
Why it is interesting
This is a practical desktop decision assistant that applies Jev to two distinct judgments: interpreting a message and assessing reply risk, then choosing among generated reply candidates. The UI keeps control with the user and the source code makes the typed request shape explicit.