How jev-chat/jev-chat-jarvis-mac works
jev-chat-jarvis is a macOS read-only chat assistant: it observes the foreground chat, analyzes new messages for intent and risk, presents ranked reply suggestions, and only inserts text after the user requests it. The main runtime coordinator is the floating HUD, which dispatches to accessibility or screen-capture adapters and coordinates analysis, generation, and display. The README also documents manual region calibration, optional detection boxes, and locally stored conversation context. Some implementation files were not sampled, so subsystem links below follow documented behavior where noted; uncertain internal wiring is omitted.
Main entry points: hud.py Floating HUD · generate.py Reply generation · registry.py App registry · ax_app.py Accessibility adapter
Components 21
Chat capture
- App registry
src/apps/registry.py - Accessibility adapter
src/apps/ax_app.py - Capture adapter
src/apps/capture_app.py - OCR perception
src/perception.py - Region calibration
src/calibration.py - Detection boxes
src/visual_fill.py - Input regions
src/input_region.py
Analysis and replies
- Local judgment
src/judge.py - Cloud judgment
src/judge_jev.py - Reply generation
src/generate.py - Reply tones
src/builtin.py
HUD and interaction
- Floating HUD
src/hud.py - Calibration UI
src/calibration_ui.py - Safe text fill
src/fill.py
Settings and context
- Context handling
src/chat_context.py - Settings window
src/settings.py - Settings controller
src/settings_config.py - User configuration
src/userconfig.py
Other components
- User
- Chat app
- Model services
Connections 32
- User → Chat app: receives messages
- User → Floating HUD: views and acts
- Floating HUD → App registry: selects adapter
- App registry → Accessibility adapter: dispatches
- App registry → Capture adapter: dispatches
- Accessibility adapter → Chat app: reads text
- Capture adapter → Chat app: captures window
- Capture adapter → OCR perception: provides image
- OCR perception → Floating HUD: returns messages
- Accessibility adapter → Floating HUD: returns messages
- Floating HUD → Local judgment: requests analysis
- Floating HUD → Cloud judgment: requests analysis
- Local judgment → Floating HUD: returns intent and risk
- Cloud judgment → Model services: sends judgment request
- Cloud judgment → Floating HUD: returns judgment
- Floating HUD → Reply generation: starts generation
- Reply generation → Reply tones: uses tone definitions
- Reply generation → Model services: requests candidates
- Reply generation → Floating HUD: returns candidates
- Floating HUD → Safe text fill: requests insertion
- Safe text fill → Chat app: checks and inserts text
- Floating HUD → Context handling: records and reads context
- Context handling → User: stores locally
- User → Calibration UI: edits regions
- Calibration UI → Region calibration: previews and saves
- Region calibration → OCR perception: supplies regions
- Floating HUD → Detection boxes: updates optional boxes
- Detection boxes → Chat app: overlays detections
- User → Settings window: configures app
- Settings window → Settings controller: edits settings
- Settings controller → User configuration: reads and writes env
- Settings controller → Context handling: updates context settings