TipTour
A macOS menu-bar companion where Jev chooses click targets from controls detected locally on screen.
What Jev decides
Which visible control to click next, which click form to use, or whether the task is complete or blocked.
In Jev mode, TipTour reads labels and locations through its local grounding engine, asks Jev for the next click target, executes the action, and validates the result until completion or a bounded stop.
- Jev role
- Choose the next compatible click action and target from controls found by the local macOS grounding engine.
- Input
- The typed goal, locally detected control labels and locations, and recent action history.
- Output
- A bounded Choice over available click actions and targets.
- Decision frequency
- Once per step in a Jev-mode desktop task, up to 12 actions.
- Fallback
- Stop on completion, no suitable target, cancellation, execution pause or failure, or the action limit; Jev mode does not fall back to Gemini automatically.
How it works
Overview
TipTour combines a local macOS grounding and action engine with two separate modes. In Jev text mode, the user types a click-based goal and Jev chooses among controls detected locally from screen and accessibility information. The app performs single, double, or right clicks, reads the interface again, and continues until Jev reports completion or none, an action fails, the user stops it, or the 12-action limit is reached.
Jev receives the typed task, detected labels and locations, and recent action history; screenshots remain local. This mode cannot interpret images, generate replacement text, or perform keyboard actions. The separate Gemini realtime mode has different capabilities and data handling and should not be attributed to Jev.
- Language
- Swift
- License
- MIT
- Author
- Milind Soni
Sources
Related projects
typesafe-computer-use
A native macOS computer-use loop built from local OCR, accessibility data, and bounded Jev actions.
What Jev decides
Which computer action, visible item, site, or off-screen accessibility control to use next.
Jev Chat Assistant
An Android conversation copilot that uses Jev to interpret chats and rank reply candidates without sending for the user.
What Jev decides
The likely intent, conversation risk, reply need, best next action, unresolved tension, and which of three candidate replies fits best.
Distill
Route coding-agent work, reasoning effort, and context retention through bounded Jev decisions.
What Jev decides
Which configured model and effort should handle a call, what evidence deserves attention, whether another attempt is useful, and what context should be retained.