Skip to content

jaffarkeikei/Mynd

Repository files navigation

Mynd

Give every AI a photographic memory of YOUR life - securely, locally, forever

The Problem: Every AI conversation starts from zero. ChatGPT doesn't remember what you discussed yesterday. Copilot doesn't know your coding style. Claude forgets your preferences. It's like having digital Alzheimer's.

The Solution: Mynd gives EVERY AI perfect memory of your context - securely, privately, forever.

What is Mynd?

Mynd is a universal memory layer for AI that automatically captures your digital context and streams it securely to any AI via Model Context Protocol (MCP). Your AIs finally remember everything about you - your decisions, preferences, history, and patterns - while your data never leaves your device.

🌟 NEW: Beautiful Web Interface

Experience Mynd through our ChatGPT-like web interface that visually demonstrates the power of AI memory:

Mynd Web Interface Demo

# Quick start the web demo
python scripts/start_web_demo.py

# Or manually:
python src/web_app.py
# Open http://localhost:8000

Key Features:

  • 🎭 Memory Toggle - Switch memory ON/OFF to see the dramatic difference
  • 🔄 Side-by-Side Comparison - Compare responses with and without memory
  • 📊 Real-time Metrics - Watch response time, relevance scores, and token usage
  • 💬 ChatGPT-like Interface - Beautiful, familiar, and intuitive
  • 🎯 Demo Mode - Pre-loaded context for instant demonstrations

See The Difference:

  1. Ask: "What was our authentication decision?"
  2. Toggle memory OFF and ask again
  3. Watch the AI go from "I don't have context" to perfect recall!

System Architecture

High-Level Architecture

graph TB
    subgraph "Data Sources"
        Browser["🌐 Browser History"]
        Files["📄 Documents & Code"]
        Clipboard["📋 Clipboard"]
        Git["🔧 Git Repositories"]
    end
    
    subgraph "Mynd Core"
        Capture["📥 Data Capture"]
        Extract["🧠 Semantic Extractor"]
        Privacy["🔒 Privacy Filter"]
        
        subgraph "Storage"
            SQLite["📊 SQLite DB<br/>(Metadata)"]
            ChromaDB["🧠 ChromaDB<br/>(Vectors)"]
        end
        
        MCP["🔗 MCP Server"]
    end
    
    subgraph "AI Clients"
        ChatGPT["💬 ChatGPT"]
        Claude["🤖 Claude"]
        Copilot["👨‍💻 GitHub Copilot"]
        AnyAI["🤖 Any AI Tool"]
    end
    
    Browser --> Capture
    Files --> Capture
    Clipboard --> Capture
    Git --> Capture
    
    Capture --> Extract
    Extract --> Privacy
    Privacy --> SQLite
    Privacy --> ChromaDB
    
    SQLite --> MCP
    ChromaDB --> MCP
    
    MCP -->|"Secure Context"| ChatGPT
    MCP -->|"Secure Context"| Claude
    MCP -->|"Secure Context"| Copilot
    MCP -->|"Secure Context"| AnyAI
    
    style Extract fill:#ff6b6b,stroke:#fff,stroke-width:3px
    style Privacy fill:#4ecdc4,stroke:#333,stroke-width:2px
    style MCP fill:#f39c12,stroke:#333,stroke-width:2px
Loading

Data Flow Process

sequenceDiagram
    participant U as User Activity
    participant C as Data Capture
    participant E as Semantic Extractor
    participant P as Privacy Filter
    participant D as Database
    participant V as Vector Store
    participant M as MCP Server
    participant A as AI Client
    
    U->>C: Browser/File/Code Activity
    C->>E: Raw Content
    E->>E: Extract Semantic Meaning
    E->>P: Semantic Events
    P->>P: Remove PII & Sensitive Data
    P->>D: Store Metadata
    P->>V: Store Embeddings
    
    Note over D,V: Local Storage Only
    
    A->>M: Request Context for Query
    M->>V: Semantic Search
    M->>D: Get Related Events
    M->>M: Compress & Optimize
    M->>A: Relevant Context (4000 tokens max)
    
    Note over M,A: MCP Protocol
Loading

The Memory Crisis (The $2.3T Problem)

Every AI interaction wastes massive time on context setup:

  • 73% of AI conversations repeat information from previous chats
  • 2.3 hours daily lost re-explaining context to AI
  • $2.3 trillion annually in global productivity loss
  • 89% of professionals frustrated with AI's goldfish memory

Real Examples:

  • "What was that API decision we made last month?" → "I don't have context"
  • "Continue our React project" → "Can you share the codebase?"
  • "Remember my coding style preferences" → "Please describe them again"

Mynd Demo Script (2 Minutes)

# The Setup (30 seconds)
"Every AI suffers from digital amnesia. Watch this..."

[User asks ChatGPT]: "What was that authentication architecture decision from last month?"
[ChatGPT]: "I don't have access to previous conversations..."

# The Magic (60 seconds)
[Install Mynd]: mynd demo
[Capture context]: "Mynd has been learning your patterns..."

[Same question to ChatGPT + Mynd]:
mynd query "authentication architecture decision"

[Result]: "You decided on JWT with refresh tokens over sessions on March 15th 
because of mobile app requirements. You were concerned about XSS attacks but 
chose client-side storage anyway because your team lacks Redis expertise."

# The Jaw-Drop (30 seconds)
"This context came from:
✅ Your browser research from 6 weeks ago
✅ Code comments you wrote in March  
✅ A design doc you saved locally
✅ All delivered securely via MCP - your data never left your machine"

Quick Start (2 Minutes to Life-Changing AI)

Component Initialization Flow

graph LR
    subgraph "Setup Process"
        Install["🔧 Install Dependencies"]
        Init["🎯 Initialize Components"]
        Demo["🎬 Create Demo Data"]
        Query["🔍 Test Query"]
    end
    
    Install --> Init
    Init --> Demo
    Demo --> Query
    
    subgraph "Components Initialized"
        DB["📊 SQLite Database"]
        Vector["🧠 Vector Store"]
        Extractor["🔍 Semantic Extractor"]
        CLI["💻 CLI Interface"]
    end
    
    Init --> DB
    Init --> Vector
    Init --> Extractor
    Init --> CLI
Loading
# Install Mynd
./install.sh  # or pip install -e .

# Set up demo data
mynd demo

# Test the magic
mynd query "authentication architecture"

# Watch AI get perfect memory of your decisions!

AgentHacks 2025 Categories

PRIMARY: Personalization & Memory

  • Learns from user activity: Continuous semantic capture
  • Evolves behavior over time: Memory graph grows and improves
  • User corrections improve system: Feedback loop for better context
  • Personal preference adaptation: Learns your patterns and style

SECONDARY: Interfaces for Human-AI Collaboration

  • Revolutionizes AI interaction: No more context re-explanation
  • Seamless collaboration: AI knows your full background
  • Natural communication: AI understands your references and history

Business Model & Market

Market Size

  • TAM: $450B (Global productivity software market)
  • SAM: $67B (AI tools and services)
  • SOM: $12B (AI productivity and memory solutions)

Revenue Model

graph TD
    Personal["🆓 Mynd Personal<br/>FREE Forever<br/>• 30-day memory<br/>• 3 data sources<br/>• Community support"] 
    
    Pro["💎 Mynd Pro<br/>$29/month<br/>• Unlimited memory<br/>• All data sources<br/>• Priority MCP access<br/>• Advanced privacy controls"]
    
    Enterprise["🏢 Mynd Enterprise<br/>$199/user/month<br/>• Team memory sharing<br/>• Compliance controls<br/>• Custom integrations<br/>• White-label deployment"]
    
    Personal --> Pro
    Pro --> Enterprise
    
    style Personal fill:#4ecdc4
    style Pro fill:#f39c12
    style Enterprise fill:#e74c3c
Loading

Security & Privacy Architecture

Privacy-First Data Flow

graph TB
    subgraph "Your Device (Secure Zone)"
        Raw["📝 Raw Data<br/>(Browser, Files, Code)"]
        PII["🔒 PII Detection<br/>(Remove Sensitive Info)"]
        LLM["🧠 Local LLM<br/>(Semantic Extraction)"]
        Encrypt["🔐 Encrypted Storage<br/>(SQLite + ChromaDB)"]
    end
    
    subgraph "External AI (Untrusted)"
        ChatGPT["💬 ChatGPT"]
        Claude["🤖 Claude"] 
        Other["🤖 Other AIs"]
    end
    
    Raw --> PII
    PII --> LLM
    LLM --> Encrypt
    
    Encrypt -->|"Semantic Context Only<br/>(No Raw Data)"| ChatGPT
    Encrypt -->|"Semantic Context Only<br/>(No Raw Data)"| Claude
    Encrypt -->|"Semantic Context Only<br/>(No Raw Data)"| Other
    
    style Raw fill:#ff6b6b,stroke:#333,stroke-width:2px
    style PII fill:#4ecdc4,stroke:#333,stroke-width:2px
    style LLM fill:#f39c12,stroke:#333,stroke-width:2px
    style Encrypt fill:#27ae60,stroke:#333,stroke-width:2px
Loading

Privacy Promise: Your raw data NEVER leaves your device. Only semantic meaning is processed, stored locally, and delivered via encrypted MCP.

Success Metrics & Validation

Technical Milestones ✅

  • Core semantic extraction engine (Local LLM + privacy filters)
  • Local encrypted storage (ChromaDB + SQLite)
  • MCP server architecture with capability tokens
  • Browser history and document capture framework
  • CLI interface with full functionality

Demo Readiness ✅

  • 2-minute live demo script prepared
  • Real context database with semantic events
  • Multiple query examples working
  • Clear before/after comparison ready

Join the Memory Revolution

Mynd isn't just a hackathon project - it's the future of AI interaction. We're building the memory layer that every AI desperately needs.

For Developers: Finally, coding AI that knows your entire project history For Knowledge Workers: AI assistants that remember every decision and context
For Everyone: The end of explaining the same thing to AI over and over


About

Mynd is a universal memory layer for AI that automatically captures your digital context and streams it securely to any AI via Model Context Protocol (MCP). Your AIs finally remember everything about you - your decisions, preferences, history, and patterns - while your data never leaves your device.

Topics

Resources

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors