How chengmatt416/lhic works
LHIC is a local-first browser and desktop automation runtime. A user submits an intent through the CLI or Desktop Control Center; the controller routes it through local Fast Path execution or optional Slow Path planning, applies risk and approval gates, executes actions, and requires verification evidence. Traces and durable workflow state support audit and recovery. The README documents additional game-training, shared-skill, MCP, and observability capabilities; source excerpts are partial, so uncertain internal wiring is omitted.
Main entry points: multi-path-task-controller.ts Intent routing · controller.ts Desktop controller · event-log.ts Trace and export · verifier-engine.ts Action verification
Components 25
User Interfaces
- CLI
apps/cli/src/main.ts - Desktop Control Center
apps/desktop/src/renderer/app.tsx - Desktop main process
apps/desktop/src/main/main.ts - Desktop controller
apps/desktop/src/main/controller.ts
Intent Control
- Intent routing
packages/controller/src/multi-path-task-controller.ts - Task context
packages/controller/src/context-engine.ts - Fast Path router
packages/controller/src/fast-path-router.ts - Slow Path planner
packages/controller/src/slow-path.ts - Model providers
packages/controller/src/openai-provider.ts
Action Execution
- Browser skills
packages/skills/src/index.ts - Playwright browser runtime
packages/browser/src/index.ts - Global desktop actions
packages/skills/src/os-bridge.ts
Security and Evidence
- Action verification
packages/verifier/src/verifier-engine.ts - Risk and approval policy
packages/security/src/risk-policy.ts - Trace redaction
packages/trace/src/redact.ts - Trace and export
packages/trace/src/event-log.ts - Workflow state
packages/memory/src/workflow-state.ts
Training and Integrations
- Game training runtime
packages/game-training/src/index.ts - Local policy trainer
packages/game-training/python/worker.py - Shared skills sync
packages/shared-skills/src/sync-service.ts - Appwrite registry
services/appwrite-shared-skills/src/main.js - MCP server
apps/mcp-server/src/index.ts - Control plane
apps/mcp-server/src/control-plane.ts
Other components
- Human user
- OTLP collector
Connections 25
- Human user → CLI: submits intent
- Human user → Desktop Control Center: submits intent
- Desktop Control Center → Desktop main process: IPC requests
- Desktop main process → Desktop controller: dispatches
- CLI → Intent routing: starts runtime
- Desktop controller → Intent routing: routes task
- Intent routing → Task context: updates context
- Intent routing → Fast Path router: routes locally
- Intent routing → Slow Path planner: plans ambiguous tasks
- Slow Path planner → Model providers: requests proposal
- Intent routing → Risk and approval policy: checks risk
- Intent routing → Browser skills: dispatches browser actions
- Browser skills → Playwright browser runtime: executes actions
- Intent routing → Global desktop actions: dispatches approved actions
- Global desktop actions → Action verification: requires verification
- Playwright browser runtime → Action verification: provides evidence
- Intent routing → Action verification: requires result verification
- Intent routing → Trace and export: records route events
- Trace and export → Trace redaction: redacts data
- Intent routing → Workflow state: recovers task state
- Game training runtime → Local policy trainer: trains policy
- Desktop controller → Game training runtime: manages training
- Shared skills sync → Appwrite registry: syncs skills
- MCP server → Control plane: serves API
- Trace and export → OTLP collector: exports traces