How the system is structured

A layered architecture that transforms model output into real-world execution, with the browser as both the host environment and a first-class execution node.

System architecture visualization
┌─────────────────────────────────────────────┐
│           AI Chat Page (Host)               │
│  ┌───────────────────────────────────────┐  │
│  │  Chrome Extension (Entry Layer)       │  │
│  │  SSE / WS / DOM Interception          │  │
│  └──────────────┬────────────────────────┘  │
│                 ↓                            │
│  ┌──────────────────────────────────────┐   │
│  │  Protocol Parser (ΩHERE / ΩBATCH)    │   │
│  └──────────────┬───────────────────────┘   │
│                 ↓                            │
│  ┌──────────────────────────────────────┐   │
│  │  Runtime & Dispatcher                │   │
│  │  ┌────────┬────────┬──────┬───────┐  │   │
│  │  │Browser │ Local  │Remote│  MCP  │  │   │
│  │  │ Bridge │ Shell  │ SSH  │ Tools │  │   │
│  │  └────────┴────────┴──────┴───────┘  │   │
│  └──────────────┬───────────────────────┘   │
│                 ↓                            │
│        Result → Conversation Stream         │
└─────────────────────────────────────────────┘

Each layer has a clear responsibility

1

Entry Layer

Chrome Extension intercepts model output via SSE/WebSocket/DOM, identifying structured commands embedded in natural language responses.

2

Protocol Layer

Parses ΩHERE, ΩBATCH, and other protocol primitives into structured action objects with parameters, preserving content fidelity.

3

Runtime & Dispatch

Routes each action to the right execution plane — browser scripts, local shell, remote SSH, or MCP tools — with sync/async/background modes.

4

Browser Execution Bridge

Leverages same-origin access to execute JavaScript in page context, read DOM, control tabs, take screenshots, and interact with web APIs directly.

5

Tool Ecosystem

50+ built-in tools spanning file I/O, media processing, database operations, cloud storage, notifications, and AI services — plus MCP for extensibility.

6

Result Feedback

Execution results are serialized and injected back into the conversation stream, enabling multi-step reasoning and self-correcting workflows.