How anshupriyan/enheduanna-meta-submission works
Enheduanna combines a local screen-memory service, a governed AI tool agent, and IGI’s screen-selection HUD. The principal workflow is: capture and index desktop activity; ask a question through the dashboard or IGI; retrieve relevant context and optionally invoke tools; stream the answer or request approval for restricted actions. The graph preserves these distinct product stages and documented privacy, visual-grounding, and timeline capabilities. Some subsystems are represented by their exact directories because their implementation files were not sampled; their internal links are omitted where evidence was unavailable.
Main entry points: engine.py Capture pipeline · orchestrator.ts Tool orchestrator · server.ts Core HTTP server · chat_window.py Companion HUD
Components 26
Agent and dashboard
- Web dashboard
app/dashboard/app.js - Core HTTP server
app/core/src/server.ts - Chat sessions
app/core/src/agent/session.ts - Tool orchestrator
app/core/src/agent/orchestrator.ts - Action approval
app/core/src/agent/approval.ts - MCP tool manager
app/core/src/mcp/manager.ts - Meta model provider
app/core/src/providers/meta.ts - Event trace bus
app/core/src/trace/bus.ts - Capture supervisor
app/core/src/capture/supervisor.ts
Desktop memory
- Capture pipeline
memvison/src/memvision/engine.py - Screen sampling
memvison/src/memvision/capture.py - Change detection
memvison/src/memvision/change_detector.py - Context extraction
memvison/src/memvision/context.py - Local OCR
memvison/src/memvision/ocr.py - Privacy filtering
memvison/src/memvision/privacy.py - Memory search store
memvison/src/memvision/storage.py - Memory MCP tools
memvison/src/memvision/mcp_server.py - Timeline viewer
memvison/src/memvision/web
IGI visual companion
- IGI launcher
igi/main.py - Selection overlay
igi/overlay/canvas.py - Spatial grounding
igi/sam/router.py - Companion HUD
igi/hud/chat_window.py - Core client
igi/bridge/core_client.py
External systems
- Meta Model API
- CLI Bridge tools
Other components
- Desktop user
Connections 30
- Desktop user → Web dashboard: asks through
- Web dashboard → Core HTTP server: sends requests
- Core HTTP server → Chat sessions: loads sessions
app/core/src/server.ts - Core HTTP server → Tool orchestrator: starts turns
app/core/src/server.ts - Tool orchestrator → Chat sessions: updates history
app/core/src/agent/orchestrator.ts - Tool orchestrator → Meta model provider: requests completions
app/core/src/agent/orchestrator.ts - Meta model provider → Meta Model API: calls models
- Tool orchestrator → MCP tool manager: executes tools
app/core/src/agent/orchestrator.ts - Tool orchestrator → Action approval: requests authorization
app/core/src/agent/orchestrator.ts - Tool orchestrator → Event trace bus: emits events
app/core/src/agent/orchestrator.ts - Core HTTP server → Capture supervisor: controls capture
app/core/src/server.ts - MCP tool manager → Memory MCP tools: connects to
app/core/src/mcp/manager.ts - MCP tool manager → CLI Bridge tools: connects to
- Capture pipeline → Screen sampling: samples screens
memvison/src/memvision/engine.py - Capture pipeline → Change detection: checks changes
memvison/src/memvision/engine.py - Capture pipeline → Context extraction: extracts context
memvison/src/memvision/engine.py - Capture pipeline → Local OCR: runs OCR
memvison/src/memvision/engine.py - Capture pipeline → Privacy filtering: redacts content
memvison/src/memvision/engine.py - Capture pipeline → Memory search store: stores captures
memvison/src/memvision/engine.py - Memory MCP tools → Memory search store: searches memory
- Desktop user → Timeline viewer: reviews timeline
- IGI launcher → Selection overlay: launches overlay
igi/main.py - IGI launcher → Companion HUD: launches HUD
igi/main.py - IGI launcher → Spatial grounding: initializes router
igi/main.py - Selection overlay → Companion HUD: passes selection
igi/main.py - Companion HUD → Spatial grounding: grounds targets
igi/hud/chat_window.py - Companion HUD → Core client: sends questions
igi/hud/chat_window.py - Core client → Core HTTP server: streams requests
igi/bridge/core_client.py - Desktop user → Selection overlay: selects screen area
- Desktop user → Companion HUD: asks visually