Summary
Make the openhuman-overlay experience reliable when the app starts, and use the floating panel to expose a clear voice on/off control alongside debug visibility for predictive autocomplete and screen intelligence so developers and power users can verify behavior without digging through logs alone.
Problem
The overlay is launched best-effort from the core when the RPC server starts (src/openhuman/overlay/process.rs), but startup behavior and the panel contents are not meeting the bar: users cannot depend on voice being easy to toggle from the overlay, and debugging autocomplete vs screen intelligence is awkward or incomplete in the UI. That blocks validation of voice flows and screen-intelligence work in real desktop sessions.
Solution (optional)
- Startup / lifecycle: Ensure the overlay launches consistently in supported dev and packaged layouts (binary discovery, ordering relative to core, failure surfacing). Adjust Tauri overlay UI as needed.
- Voice: Add or fix a visible toggle that turns voice capture / dictation pipeline on and off, with state that matches the actual audio path.
- Debug: In the same overlay surface, expose actionable debug for autocomplete (e.g. context/suggestions path) and screen intelligence (feature status, last events, errors) without requiring separate tools—scoped so normal users are not overwhelmed (e.g. dev-only section or compact indicators).
Coordinate with screen intelligence / autocomplete RPC and macOS-only constraints documented in the engine (src/openhuman/screen_intelligence/).
Acceptance criteria
Related
Summary
Make the
openhuman-overlayexperience reliable when the app starts, and use the floating panel to expose a clear voice on/off control alongside debug visibility for predictive autocomplete and screen intelligence so developers and power users can verify behavior without digging through logs alone.Problem
The overlay is launched best-effort from the core when the RPC server starts (
src/openhuman/overlay/process.rs), but startup behavior and the panel contents are not meeting the bar: users cannot depend on voice being easy to toggle from the overlay, and debugging autocomplete vs screen intelligence is awkward or incomplete in the UI. That blocks validation of voice flows and screen-intelligence work in real desktop sessions.Solution (optional)
Coordinate with screen intelligence / autocomplete RPC and macOS-only constraints documented in the engine (
src/openhuman/screen_intelligence/).Acceptance criteria
Related
src/openhuman/overlay/process.rssrc/openhuman/screen_intelligence/