How fatihmakes/mark-li works
MARK LIV is a cross-platform voice-and-visual personal assistant. A user speaks or types in the desktop UI; the live session handles conversation, invokes discovered actions or plugins for tasks, and returns spoken and visual results. Supporting branches provide persistent memory, visual awareness, the animated avatar, and remote phone control. The README documents broad capabilities, but most runtime wiring is unsampled, so links are limited to documented behavior and visible source evidence.
Main entry points: main.py Live Session · ui.py Desktop UI · actions Assistant Actions · action_loader.py Action Discovery
Components 18
Interaction
- Desktop UI
ui.py
Live Assistant
- Live Session
main.py - Audio and Speech
core - Screen and Vision
actions/screen_processor.py
Skills
- Action Discovery
core/action_loader.py - Assistant Actions
actions - User Plugins
plugins - Planning LLM
core/llm_client.py - Web Search
actions/web_search.py
Memory and Settings
- Persistent Memory
memory/memory_manager.py - Configuration
memory/config_manager.py - Undo
core/undo.py
Displays and Remote
- Remote Dashboard
dashboard/server.py - Animated Avatar
core/avatar.py - Avatar Geometry
core/avatar_mesh.py
Other components
- User
- Gemini Live API
- External Services
Connections 21
- User → Desktop UI: speaks or types
- Desktop UI → Live Session: provides input
- Live Session → Audio and Speech: streams audio
- Audio and Speech → Gemini Live API: sends and receives
- Live Session → Action Discovery: discovers tools
- Action Discovery → Assistant Actions: loads actions
- Action Discovery → User Plugins: loads plugins
- Live Session → Assistant Actions: dispatches tools
- Live Session → Persistent Memory: reads and updates
- Live Session → Configuration: reads settings
- Live Session → Undo: dispatches undo
- Live Session → Screen and Vision: requests screen processing
- Live Session → Planning LLM: requests planning
- Live Session → Desktop UI: updates display
- Live Session → Animated Avatar: provides audio state
- Animated Avatar → Avatar Geometry: gets head geometry
- Desktop UI → Remote Dashboard: starts remote access
- Remote Dashboard → Live Session: queues commands
- Assistant Actions → Web Search: dispatches search
- Assistant Actions → External Services: uses integrations
- Gemini Live API → External Services: uses live service