Skip to content

ARCHITECTURE.md

Latest commit

 

History

History
94 lines (75 loc) · 3.06 KB

ARCHITECTURE.md

File metadata and controls

94 lines (75 loc) · 3.06 KB

ReasonKit Web Architecture

Web Sensing Layer (Rust)

The Web Sensing Layer provides high-performance browser automation and content extraction for AI reasoning systems.

Architecture Diagram

flowchart LR
    subgraph WEB_SENSING_LAYER_RUST["WEB SENSING LAYER (RUST)"]
        A[AI Agent]
        B[MCP Server]
        C[Browser Controller<br/>CDP]

        A --> B --> C

        subgraph PIPELINE[""]
            D[Capture<br/>Screenshots / PDFs / HTML]
            E[Extraction<br/>Content + Metadata]
        end

        C --> D --> E
    end

    E -.->|delegates to| F[reasonkit-mem<br/>Triangulation]
Loading

Note: Content extraction happens in reasonkit-web. Multi-source triangulation and verification are handled by reasonkit-mem via the PyO3 bridge (see TRIANGULATION_PROTOCOL_V2.md).

Component Flow

  1. AI Agent → Sends requests via Model Context Protocol
  2. MCP Server → Receives and routes requests to browser controller
  3. Browser Controller (CDP) → Controls headless Chromium via Chrome DevTools Protocol
  4. Capture → Captures screenshots, PDFs, and HTML from web pages
  5. Extraction → Extracts main content, metadata, and structured data
  6. Triangulation → (Delegated to reasonkit-mem) Verifies claims across multiple sources

Technology Stack

Component Technology Purpose
Browser ChromiumOxide Async Rust CDP client
MCP Server mcp-sdk-rs AI agent integration
Runtime Tokio Async runtime
Serialization Serde JSON handling

Module Structure

reasonkit-web/
├── browser/          # Browser automation (CDP)
│   ├── controller.rs # Main browser controller
│   ├── capture.rs    # Screenshot, PDF, HTML capture
│   ├── navigation.rs # Page navigation
│   └── stealth.rs    # Anti-detection measures
├── extraction/       # Content extraction
│   ├── content.rs    # Main content extraction
│   ├── links.rs      # Link extraction
│   └── metadata.rs   # Metadata extraction
└── mcp/              # MCP server implementation
    ├── server.rs     # MCP server
    ├── tools.rs      # MCP tools (browser operations)
    └── types.rs      # MCP type definitions

Data Flow

AI Agent Request
    ↓
MCP Server (JSON-RPC 2.0)
    ↓
Browser Controller (CDP)
    ↓
┌─────────────┬─────────────┐
│   Capture   │  Extraction │
└──────┬──────┴──────┬───────┘
       │             │
       ▼             ▼
Screenshots    Content + Metadata
PDFs, HTML    Links, Structured Data
       │             │
       └──────┬──────┘
              ↓
      reasonkit-mem (Triangulation)
              ↓
      Verified Facts

"Designed, Not Dreamed" | https://reasonkit.sh