How nikeshsundar/argus works
Argus is a Windows desktop assistant: users invoke it from a hotkey or overlay, ask about the current screen, or delegate desktop work. The main process coordinates providers, screen capture, action execution, approvals and persistence; renderer and preload code provide the interaction surface. Documented branches include voice, screen memory, saved workflows, Teach Mode, SOPs and model selection. Several domain modules are indexed but unsampled, so their internal call paths are not inferred.
Main entry points: index.ts Main process · agentLoop.ts Agent loop · main.ts Overlay UI · overlayWindow.ts Overlay window
Components 23
Desktop interaction
- Hotkey
src/main/hotkey.ts - Main process
src/main/index.ts - Preload bridge
src/preload/index.ts - Overlay UI
src/renderer/src/main.ts - Overlay window
src/main/overlayWindow.ts - Voice input
src/renderer/src/voice.ts
AI and execution
- Agent loop
src/main/agentLoop.ts - Screen capture
src/main/screenshot.ts - Gemini agent
src/main/providers/geminiAgent.ts - Talk providers
src/main/providers - Gemini client
src/main/providers/geminiClient.ts - Desktop actions
src/main/inputSim.ts
Assistant capabilities
- Screen memory
src/main/screenMemory.ts - Recipes and research
src/main/recipes.ts - Teach mode
src/main/teachLoop.ts - SOP runner
src/main/sopRunner.ts - Saved workflows
src/main/workflowStore.ts
State and safeguards
- Settings
src/main/settingsStore.ts - Chat history
src/main/history.ts - Action safety
src/shared/safety.ts - Agent limits
src/shared/limits.ts - Gemini key pool
src/main/geminiKeys.ts
Other components
- User
Connections 27
- User → Hotkey: invokes
README.md - Hotkey → Main process: signals
src/main/index.ts - Main process → Agent loop: dispatches
src/main/index.ts - Main process → SOP runner: dispatches
src/main/index.ts - Main process → Recipes and research: dispatches
src/main/index.ts - Main process → Saved workflows: uses
src/main/index.ts - Main process → Settings: loads
src/main/index.ts - Main process → Talk providers: selects
src/main/providers/index.ts - Main process → Chat history: persists
src/main/index.ts - Main process → Overlay window: controls
src/main/overlayWindow.ts - Overlay window → Preload bridge: loads
src/main/overlayWindow.ts - Preload bridge → Overlay UI: bridges
src/preload/index.ts - Overlay UI → Voice input: uses
src/renderer/src/main.ts - Agent loop → Screen capture: captures
src/main/agentLoop.ts - Agent loop → Gemini agent: creates session
src/main/agentLoop.ts - Agent loop → Desktop actions: executes actions
src/main/agentLoop.ts - Agent loop → Overlay window: reports progress
src/main/agentLoop.ts - Agent loop → Settings: loads limits
src/main/agentLoop.ts - Agent loop → Action safety: checks risk
src/main/agentLoop.ts - Agent loop → Agent limits: applies constraints
src/main/agentLoop.ts - Talk providers → Gemini client: uses
src/main/providers/geminiClient.ts - Gemini client → Gemini key pool: takes keys
src/main/providers/geminiClient.ts - Screen memory → Overlay UI: shown in Talk Mode
README.md - Recipes and research → Gemini client: uses
src/main/recipes.ts - Teach mode → Overlay window: reports steps
- SOP runner → Agent loop: runs screen steps
src/main/sopRunner.ts - Saved workflows → Overlay UI: offered to user
README.md