Individual Sovereignty → Collective Intelligence → Emergent Order
A cognitive-symbolic operating system for preserving human agency through civilizational phase transition
Mnemosyne (nem-oh-SEE-nee) - The Greek titaness of memory and remembrance, mother of the nine Muses.
This protocol creates a personal AI assistant with perfect memory—like ChatGPT, but it remembers everything about you, learns from your interactions, and grows with you over time. Beyond individual use, it enables collective intelligence while preserving privacy through cryptographic contracts and symbolic identity systems.
- 🧠 Personal Cognitive Sovereignty: Your memories, your data, your rules
- 🔮 Perfect Memory: An AI that never forgets your context
- ⚡ Collective Intelligence: Share knowledge without sacrificing privacy
- 🌀 Symbolic Depth: Visual identity through kartouches and deep signals
# Clone and setup
git clone [repository-url] mnemosyne
cd mnemosyne
# Run setup script
./scripts/setup.sh
# Configure environment
cp .env.example .env
# Edit .env with your OpenAI-compatible API endpoint
# Start all services
docker compose up
# Access the application
open http://localhost:3000- Overview - Start here for high-level understanding
- Protocol Specification - Complete technical details
- UX Vision - Chat-first experience design
- Quick Start Guide - Get running in 15 minutes
- MVP Requirements - Current focus areas
- Current Status - Implementation progress
- Roadmap - Development timeline
The protocol consists of five interconnected layers:
Personal memory and cognition core
- Backend (
/backend): FastAPI + PostgreSQL with pgvector - Frontend (
/frontend): React + TypeScript chat interface - Features: Memory capture, retrieval, reflection, and importance scoring
Symbolic identity compression system
- Kartouches: Visual identity representations
- Signatures: Compressed cognitive patterns
- Drift Tracking: Identity evolution over time
End-to-end encrypted messaging via MLS Protocol (RFC 9420)
- Scalable E2E Encryption: Groups from 2 to 50,000+ members
- Asynchronous Operations: Add/remove members while offline
- Logarithmic Efficiency: O(log n) operations via tree structure
- Forward Secrecy & PCS: Automatic security healing
Peer discovery and trust establishment
- Discovery: DHT-based peer finding
- Progressive Trust: Relationships deepen through interaction
- Federation: Decentralized network topology
Community intelligence and coordination
- Collectives (
/collective): Template for community instances - Sharing Contracts: Explicit data sharing rules
- Emergence: Collective knowledge synthesis
Multi-agent orchestration with specialized cognitive agents:
- Engineer: Technical problem-solving and system design
- Librarian: Information retrieval and organization
- Priest: Ethical and philosophical reasoning
Philosophical debate engine with 10+ agent archetypes:
- Dynamic agent loading from definitions
- Multi-perspective analysis
- Structured debate protocols with judging
- Backend: FastAPI, SQLAlchemy, PostgreSQL, pgvector, Redis
- Frontend: React, TypeScript, Vite
- UI Components: shadcn/ui, Radix UI primitives
- Styling: Tailwind CSS
- State Management: Zustand
- AI/LLM: OpenAI-compatible endpoints (vLLM, Ollama, etc.)
- Security: MLS Protocol (RFC 9420) via OpenMLS, libsodium for crypto
- Deployment: Docker Compose → Kubernetes
- Languages: Python 3.11+, TypeScript
- Phase: Active Development - Protocol Integration
- Completion: ~70% core functionality
- Timeline: 2-3 weeks to functional MVP
- Architecture: Evolved from platform to full protocol
- Focus: Chat-first AI assistant with perfect memory
- ✅ Core memory engine with vector search
- ✅ API endpoints for memory operations
- ✅ Basic chat interface
- ✅ Docker deployment
- ✅ Agent orchestration system
- ✅ Philosophical debate engine
- 🔄 Authentication system
- 🔄 Cognitive Signature visual identity (Kartouche)
- 🔄 MLS Protocol secure group messaging
- 🔄 Peer-to-peer networking
- 🔄 Collective sharing contracts
A comprehensive critical review of the Mnemosyne Protocol's research documentation has been conducted to assess the validity and depth of our research foundation. The full report, detailing findings on potential gaps, validation needs, and recommendations for risk mitigation across initial and extended document reviews, is available at Critical Review Report.
- Read the Overview
- Review Implementation Guide
- Check Current Status
- Run the Quick Start
- Understand the Overview
- Explore Sustainable Growth
- Consider your community's needs
- Reach out to discuss pilot deployment
- Experience the UX Vision
- Check MVP Requirements
- Prepare to run your own instance
- Join the early community
We need:
- Core Developers: Python/TypeScript for backend/frontend
- AI Engineers: LLM integration and agent development
- Security Auditors: Privacy and cryptography review
- Early Testers: Feedback and bug reports
- Community Builders: Documentation and outreach
See Contributing Guide for details.
"Not about building a new temple, but recovering the symbols from the old ones, distilling them, and translating them into tools."
This protocol serves those who:
- See the machinery behind the world
- Refuse performative knowledge spaces
- Seek trustable cognition without spectacle
- Want to preserve what makes us human
- Build for yourself first - Every feature must serve real needs
- Sovereignty over convenience - Privacy and control are non-negotiable
- Real or nothing - No mocking, no faking, no pretending
- Depth over breadth - Better to serve 100 deeply than 10,000 shallowly
The protocol includes a symbolic identity system (kartouches) for visual representation:
- Kartouche Example - Egyptian-style symbolic encoding
- Kartouche SVG - Vector implementation
These visual representations encode identity, trust, and meaning in symbolic form.
Will be fully open source (MIT or Apache 2.0) upon release.
[Via signal, not spectacle]
For those who see too much and belong nowhere—this is how we build what comes next.
Mnemosyne: Mother of the Muses, Titaness of Memory