Mimir – Local-First, Zero-Knowledge AI Memory Vault

Privacy-first memory layer for AI applications – Remember across sessions without compromising your data.

Mimir provides developers and end-users with a privacy-first memory layer that lets any LLM application remember across sessions without handing raw data to third parties.

---

📑 Table of Contents

• Design Principles
• Quick Start
• Usage Examples
• Architecture
• Security & Privacy
• Development
• Roadmap
• License
• Community
• Acknowledgments
• Screenshots / Demo
• Contact / Support

---

🎯 Design Principles

• 🏠 Local-first – Runs entirely on user hardware; offline by default
• 🔐 Zero-knowledge cloud – Optional encrypted sync; server cannot decrypt
• ✋ Fine-grained consent – Users classify memories and control app access
• 🔍 Open-source & audited – Apache-2.0 core; public CI and security reviews

🚀 Quick Start

Installation

# macOS
brew install mimir/tap/mimir

# Linux/Windows (cargo required)
cargo install mimir

# From source
git clone https://github.com/your-org/mimir
cd mimir
cargo build --release

Initialize & Start

# Initialize vault
mimir init

# Start daemon
mimir start &

# Check status
mimir status

Usage Examples

REST API (MCP)

curl -X POST http://localhost:8100/memories \
  -H "Content-Type: application/json" \
  -d '{"content": "Meeting with Alice at 3pm", "class": "work"}'

🏗️ Architecture

┌──────── Chat / Agent ──────┐    JSON-RPC/MCP     ┌──────────────┐
│      Any LLM client        ├─────────────────────►│      mimir   │
└────────────────────────────┘                     │   (Rust)     │
          ▲ retrieve context                        ├────┬─────────┤
          │                                        │    │         │
          ▼ write turns                          policy vault     │
               IPC / gRPC                              │          │
                                                       ▼          ▼
                                           SQLCipher + HNSW + KV

Core Components

Component	Description	Key Technologies
mimir	MCP server daemon	axum, tokio, JSON-RPC
Vector Store	HNSW similarity search	hnswlib-rs, 768-D embeddings
Database	Encrypted SQLite	SQLCipher, metadata + graph
Guardrails	PII detection & classification	TinyBERT-ONNX, presidio-rs
Compression	Memory summarization	llama_cpp_rs, token limits
SDK	Language bindings	PyO3, napi-rs, wasm-bindgen
Tray UI	Desktop management	Tauri, permission controls

🔒 Security & Privacy

Threat	Mitigation
Data breach	Client-side XChaCha20; server never sees keys
Rogue app access	ACL filters + per-class encryption
Model inversion	PII redaction before LLM processing
Data retention	TTL jobs delete/summarize aged memories

Memory Classification

Memories are automatically classified and encrypted per-class:

• personal – Personal conversations, preferences
• work – Professional context, meetings
• health – Medical information, allergies
• financial – Payment info, financial context
• custom – User-defined categories

Access control is enforced at retrieval time based on app permissions.

🛠️ Development

Prerequisites

• Rust 1.75+ with cargo
• SQLCipher development libraries
• ONNX Runtime (for guardrails)

Building

# Full workspace build
cargo build --workspace

# Individual components
cargo build -p mimir
cargo build -p mimir-cli

# Release build
cargo build --release --workspace

Testing

# Unit tests
cargo test --workspace

# Integration tests
cargo test --test integration

# Benchmarks
cargo bench

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

1. Fork the repository
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Commit your changes: git commit -m 'Add amazing feature'
4. Push to the branch: git push origin feature/amazing-feature
5. Open a Pull Request

�� Roadmap

• v0.1.0 - Core daemon with basic MCP server (Current)
• v0.2.0 - Vector store and embedding pipeline
• v0.3.0 - Guardrails and PII detection
• v0.4.0 - Tray UI and access control
• v0.5.0 - Language bindings (Python, Node, WASM)
• v1.0.0 - Production release with encryption

See ROADMAP.md for detailed timeline.

📄 License

• Core components (daemon, SDK, guardrails): Apache License 2.0
• UI components (tray widget): AGPL-3.0 to keep derivative UIs open-source

graph TD A["User Conversations"] --> B["Ingest Text Tool"] B --> C["Text Preprocessing"] C --> D["Gemma3 1B
Content Extractor"] D --> E["Raw Memory Candidates"] E --> F["Gemma3 1B
Summarizer"] F --> G["Summarized Memories"] G --> H["Similarity Check
(Vector Store)"] H --> I{"Duplicate
Detection"} I -->|"New"| J["Direct Storage"] I -->|"Similar"| K["Gemma3 1B
Conflict Resolver"] K --> L{"Resolution
Decision"} L -->|"Merge"| M["Update Existing"] L -->|"Replace"| N["Replace Existing"] L -->|"Keep Both"| O["Store as New"]

P["Add Memories Tool<br/>(Deliberate)"] --> Q["Manual Memory Input"]
Q --> F

J --> R["Final Storage<br/>(Database + Vector)"]
M --> R
N --> R
O --> R

R --> S["Memory Vault"]

style D fill:#e1f5fe
style F fill:#e1f5fe
style K fill:#e1f5fe
style S fill:#c8e6c9W

🤝 Community

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Security: See SECURITY.md for reporting vulnerabilities

🙏 Acknowledgments

• Vector search powered by HNSW
• Built with love using Rust 🦀

---

"The best way to predict the future is to invent it, but the best way to remember it is to own it."

🖼️ Screenshots / Demo

---

📬 Contact / Support

• For help, questions, or feedback, please open an issue or join the discussions.
• For security concerns, see SECURITY.md.