MineContext Architecture Overview

System Architecture

MineContext is designed as a modular, event-driven system with clear separation of concerns. The architecture follows a layered approach with well-defined interfaces between components.

System Components

Core Architecture Layers

┌──────────────────────────────────────────────────────────┐
│                    Business Logic Layer                  │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐  │
│  │   Capture    │  │  Processor   │  │  Consumption │  │
│  │   Manager    │  │   Manager    │  │    Manager   │  │
│  └──────────────┘  └──────────────┘  └──────────────┘  │
└──────────────────────────────────────────────────────────┘
                           ↓
┌──────────────────────────────────────────────────────────┐
│                    Processing Layer                      │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐  │
│  │    Chunker   │  │   Processor  │  │    Merger    │  │
│  └──────────────┘  └──────────────┘  └──────────────┘  │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐  │
│  │Entity Analyzer│ │  Vectorizer  │  │  Knowledge   │  │
│  │              │  │              │  │  Extractor   │  │
│  └──────────────┘  └──────────────┘  └──────────────┘  │
└──────────────────────────────────────────────────────────┘
                           ↓
┌──────────────────────────────────────────────────────────┐
│                     Storage Layer                        │
│  ┌──────────────┐  ┌──────────────┐  │
│  │    SQLite    │  │   ChromaDB   │  │
│  └──────────────┘  └──────────────┘  │
└──────────────────────────────────────────────────────────┘
                           ↓
┌──────────────────────────────────────────────────────────┐
│                        LLM Layer                         │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐  │
│  │    OpenAI    │  │    Doubao    │  │  Vectorization│ │
│  │              │  │              │  │     Models   │  │
│  └──────────────┘  └──────────────┘  └──────────────┘  │
└──────────────────────────────────────────────────────────┘

System Flow

The MineContext system flow is designed as a highly modular and extensible architecture, divided into five core components: context capture, processing, storage, services/tools, and consumption.

1. Context Capture

• Sources:
  • Local Files: Monitor and capture locally created or modified files
  • Data Stream Subscriptions: Subscribe and receive information from various data sources (e.g., RSS, APIs)
  • User Interactions: Including chat logs, screenshots, and AI conversations
  • Custom Methods: Support user-defined capture methods

2. Context Processing

• Document Processing:
  • Structured Documents: Perform local segmentation
  • Unstructured Documents: Slice through content understanding
• Multimodal Understanding:
  • Comprehensive understanding of screenshots, chat logs, etc.
  • Retrieve recent historical content and context information
• Entity Extraction & Updates: Identify and update entities from processed content
• Knowledge Extraction:
  • Denoise, merge, and classify information
  • Form objective knowledge, entity profiles, activity records, and process records
• Vectorization: Transform extracted knowledge into vector form for storage and retrieval
• Time-driven Compression: Compress context and retrieve associated events and similar knowledge

3. Context Storage

• Vector Databases:
  • Objective Knowledge Vector Store: Store facts and general knowledge
  • Event Records Vector Store: Store information related to specific events
  • Entity Relations Vector Store: Store relationships between entities
  • Others: Store other types of processed data

4. Context Services/Tools

• Query Processing:
  • Intent Recognition & Rewriting: Understand the true intent of user queries, rewrite and decompose them
• Information Organization:
  • Relationship Networks: Retrieve entity profiles and their relationship networks in events
  • Timeline Views: Generate timeline views based on semantic matching of time and actions
• Retrieval & Aggregation:
  • Knowledge Aggregation: Retrieve and aggregate fact-oriented and reusable knowledge
  • Bidirectional Query: Support bidirectional queries from time to event or event to time
  • Hybrid Search: Support cross-type semantic and keyword hybrid search with filtering, sorting, grouping, and weight decay features

5. Context Consumption

• MCP Server: Provide services through Model Context Protocol (MCP)
• Application Layer Consumption: Provide processed and organized context to upper-layer applications

Data Flow

┌─────────────────────────────────────────────────────────────┐
│                      Data Input Sources                     │
│  ┌────────────┐  ┌────────────┐  ┌────────────┐  ┌────────────┐ │
│  │Local Files │  │Data Stream │  │User        │  │Custom     │ │
│  │            │  │Subscription│  │Interaction │  │Sources    │ │
│  └────────────┘  └────────────┘  └────────────┘  └────────────┘ │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                     Capture Manager                          │
│      (Monitor, Retrieve, Receive, Custom Capture Methods)    │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                    Processor Manager                         │
│  ┌──────────────────────┐  ┌────────────────────────────┐ │
│  │  Document Processing │  │    Multimodal Understanding│ │
│  │  • Structured doc    │  │  • Screenshot/chat         │ │
│  │    segmentation      │  │    understanding           │ │
│  │  • Unstructured doc  │  │  • Historical content      │ │
│  │    slicing           │  │    retrieval               │ │
│  └──────────────────────┘  └────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                 Knowledge Extraction Processing              │
│  ┌──────────────────────┐  ┌────────────────────────────┐ │
│  │Entity Extraction &   │  │    Information Extraction  │ │
│  │      Updates         │  │  • Denoise/Merge/Classify │ │
│  │  • Entity recognition│  │  • Objective knowledge/    │ │
│  │  • Relation extraction│ │    Entity profiles         │ │
│  │  • Entity profile    │  │  • Activity/Process        │ │
│  │    updates           │  │    records                 │ │
│  └──────────────────────┘  └────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                  Vectorization & Compression                 │
│  ┌──────────────────────┐  ┌────────────────────────────┐ │
│  │ Vectorization Process│  │   Time-driven Compression │ │
│  │  • Text vectorization│  │  • Context compression     │ │
│  │  • Multimodal        │  │  • Related event retrieval│ │
│  │    vectorization     │  │  • Similar knowledge       │ │
│  │  • Feature extraction│  │    aggregation             │ │
│  └──────────────────────┘  └────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                      Storage Manager                         │
│  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐ │
│  │  Objective   │  │    Event     │  │    Entity    │ │
│  │  Knowledge   │  │   Records    │  │  Relations   │ │
│  │(Vector Store)│  │(Vector Store)│  │(Vector Store)│ │
│  └──────────────┘  └──────────────┘  └──────────────┘ │
│                      SQLite + ChromaDB                      │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                   Query & Retrieval Services                 │
│  ┌──────────────────────┐  ┌────────────────────────────┐ │
│  │   Query Processing   │  │    Retrieval Aggregation   │ │
│  │  • Intent recognition│  │  • Knowledge aggregation   │ │
│  │  • Query rewriting/  │  │  • Bidirectional query     │ │
│  │    decomposition     │  │  • Hybrid search           │ │
│  │  • Parameter         │  │                            │ │
│  │    extraction        │  │                            │ │
│  └──────────────────────┘  └────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                 Information Organization Service             │
│  ┌──────────────────────┐  ┌────────────────────────────┐ │
│  │Relationship Network  │  │     Timeline Generation    │ │
│  │    Construction      │  │  • Time sequence matching │ │
│  │  • Entity profiles   │  │  • Action semantic        │ │
│  │  • Event relationship│  │    matching                │ │
│  │    networks          │  │  • Timeline views          │ │
│  │  • Entity relation   │  │                            │ │
│  │    graphs            │  │                            │ │
│  └──────────────────────┘  └────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                     Consumption Interface                    │
│  ┌──────────────────────┐  ┌────────────────────────────┐ │
│  │     MCP Server       │  │   Application Layer API    │ │
│  │                      │  │                            │ │
│  └──────────────────────┘  └────────────────────────────┘ │
└─────────────────────────────────────────────────────────────┘

Extension Points

1. Adding New Context Sources

• Implement CaptureInterface
• Register in CaptureManager

2. Adding New Processing Strategies

• Implement BaseProcessor
• Register in ProcessorFactory

3. Adding New Storage Backends

• Implement BaseStorage
• Update configuration options

4. Adding New LLM Providers

• Implement LLM client interface
• Update configuration handling

Configuration Management

The system uses a layered configuration approach:

1. Default Configuration: Built-in default values
2. File Configuration: config.yaml overrides
3. Environment Variables: Override file configuration
4. Runtime Configuration: API-based configuration updates

Security Considerations

1. API Security

• Token-based authentication
• Rate limiting
• Input validation

2. Data Privacy

• Local storage
• Configurable data retention
• Secure API key management
• No storage of private data

Deployment Architecture

Standalone Deployment

┌─────────────┐
│  MineContext    │
│  Application    │
│                 │
│  - Web Server   │
│  - Processing   │
│  - Storage      │
└─────────────────┘

Distributed Deployment

┌─────────────────┐     ┌─────────────────┐
│   Web Server    │────▶│ Processing Workers│
│   (Frontend)    │     │                 │
└─────────────────┘     └─────────────────┘
         │                       │
         └───────┬───────────────┘
                 ▼
         ┌───────────────┐
         │ Storage Backend│
         └───────────────┘