MemWyre

Your Second Brain, Supercharged.

MemWyre is a universal memory layer for AI—designed to capture, organize, and retrieve your knowledge across tools, conversations, and large language models.

Instead of losing context every time you switch between ChatGPT, Gemini, Claude, editors, or agents, MemWyre becomes the persistent brain that follows you everywhere.

Demo

The Problem

Today’s AI tools are powerful—but stateless. Every chat starts from zero Important insights get lost in conversations Knowledge is scattered across PDFs, chats, docs, videos, and web pages Each LLM lives in its own silo Users are forced to re-explain themselves endlessly Existing tools solve storage or note-taking, but not memory continuity for AI. AI forgot. Again.

What MemWyre Solves

MemWyre introduces a shared, intelligent memory system that sits outside any single LLM and works with all of them.

It enables: Long-term memory for AI workflows Cross-LLM knowledge reuse Contextual retrieval grounded in your own data A clean separation between thinking (LLMs) and remembering (MemWyre) MemWyre doesn’t replace AI tools—it augments them.

Key Features

Unified Memory Vault

Store everything in one place:

Chat outputs Documents (PDF, DOC, Markdown) Web pages & research Notes, ideas, decisions Agent outputs

All content is automatically processed, chunked, embedded, and indexed for semantic retrieval.

Smart Memory (Not Just Storage)

MemWyre goes beyond dumping files: Semantic chunking Auto-generated embeddings Metadata & source tracking Context-aware retrieval Inbox-based memory approval flow Memory becomes queryable intelligence, not dead notes.

Contextual Retrieval Engine

Ask questions and get answers grounded in your own memory, not hallucinations.

Vector search Top-K relevance ranking Source-linked context Optional summaries

Works as a standalone search or as context injected into LLM prompts.

LLM-Agnostic by Design

MemWyre is not tied to any single model. It integrates via: API-based LLM connectors MCP servers (for editors & agents) Browser extensions Prompt-level injection (for restricted platforms)

Your memory works across ChatGPT, Gemini, Claude, local models, and future agents.

Intelligent Memory Inbox

Not all memory should be saved blindly. MemWyre introduces an Inbox model Review before committing to long-term memory Auto-approve trusted sources Manual control where it matters Prevents memory pollution

This keeps the system useful, not noisy.

Multi-Source Ingestion

Bring knowledge from everywhere:

File uploads Browser extension Web scraping YouTube transcripts Agent outputs Code editor integrations

All paths lead to the same memory engine.

Quick Start

Prerequisites

• Python 3.10+
• Node.js 18+
• Chrome/Edge browser (for extension)

Backend Setup

cd backend
python -m venv venv
venv\Scripts\activate  # Windows
# source venv/bin/activate  # Linux/Mac

pip install -r requirements.txt
python -m uvicorn app.main:app --reload

Backend runs at http://localhost:8000

Worker Setup (Background Tasks)

Requires a running Redis instance.

cd backend
venv\Scripts\activate
# Windows (requires -P solo)
celery -A app.celery_app worker --loglevel=info -P solo

Frontend Setup

cd frontend
npm install
npm run dev

Frontend runs at http://localhost:5173

First-Time Setup

1. Navigate to http://localhost:5173
2. Register a new account
3. Login with your credentials
4. Upload documents or create memories via the dashboard

Usage Guide

Tier 1: Prompt Generator

1. Click "Generate Prompt" on the dashboard
2. Enter your query (e.g., "Summarize my project notes")
3. Select a template and adjust context size
4. Click "Generate Prompt"
5. Copy and paste into any LLM

Tier 2: Browser Extension

Installation

1. Open Chrome and go to chrome://extensions/
2. Enable Developer mode
3. Click Load unpacked
4. Select extension folder from the project root

Usage

1. Click the extension icon
2. Go to Settings in Brain Vault web app
3. Click "Copy Extension Token"
4. Paste token into extension popup
5. Use the popup to generate prompts or let it auto-inject on supported sites

Tier 3: MCP Server

For Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "brain-vault": {
      "command": "python",
      "args": ["C:/Users/himan/OneDrive/Documents/brain_vault/backend/mcp_server.py"]
    }
  }
}

For VS Code / Cursor

Configure your MCP client to run:

python backend/mcp_server.py

Available Tools

• search_memory(query, top_k) - Search your knowledge base
• save_memory(text, tags) - Save new memories
• get_document(doc_id) - Retrieve full document content

Architecture

brain_vault/
├── backend/           # FastAPI server
│   ├── app/
│   │   ├── routers/   # API endpoints
│   │   ├── models/    # Database models
│   │   ├── services/  # Business logic
│   │   └── core/      # Config & security
│   ├── mcp_server.py  # MCP server
│   └── requirements.txt
├── frontend/          # Vue.js app
│   ├── src/
│   │   ├── views/     # Pages
│   │   ├── components/# UI components
│   │   └── services/  # API client
│   └── package.json
└── extension/         # Chrome extension
    ├── manifest.json
    ├── background.js
    ├── content.js
    └── popup.html

Configuration

API Keys

Set your LLM API keys in Settings:

• OpenAI API Key (for GPT models)
• Gemini API Key (for Gemini models)

Keys are stored locally in browser localStorage.

Database

SQLite database is created automatically at backend/brain_vault.db.

Vector Store

ChromaDB stores embeddings at backend/chroma_db/.

Testing

Test MCP Server

cd backend
venv\Scripts\activate
python test_mcp_live.py

Test Prompt Generation

1. Navigate to /prompts in the web app
2. Enter a test query
3. Verify prompt is generated with context

API Documentation

Once the backend is running, visit:

• Swagger UI: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc

Tech Stack

Backend:

• FastAPI
• SQLAlchemy
• ChromaDB
• LangChain
• MCP (Model Context Protocol)

Frontend:

• Vue 3
• Vite
• Tailwind CSS
• Monaco Editor

Extension:

• Manifest V3
• Chrome Extension APIs

Contributing

This is a personal project, but suggestions are welcome via issues.

License

MIT License - feel free to use and modify.

Acknowledgments

• Built with assistance from Antigravity AI
• MCP specification by Anthropic
• Inspired by personal knowledge management best practices