I discovered a problem that existing AI memory systems don't solve

Memory Firewall is still under active development, and I'm looking for developers, researchers, and AI enthusiasts who find this problem interesting.

GitHub Repository:

NitheshK4 / memory-firewall

A secure gateway and policy engine for AI agent memory. Intercepts, inspects, and sanitizes read/write loops to prevent prompt injection and memory poisoning.

Memory Firewall

Memory Firewall is a runnable MVP for defending long-term memory in AI agents It intercepts memory writes and memory retrievals, scores them for risk, records provenance, checks contradictions, and quarantines suspicious content before it can silently corrupt future agent behavior.

What is included

• FastAPI service for memory ingestion, retrieval, review, and health checks
• LangGraph-based write and read firewall flows
• Typed schemas for claims, provenance, verdicts, and stored memories
• In-memory repository for zero-friction local demos
• Docker Compose scaffold for Postgres and Neo4j expansion
• Streamlit dashboard for quarantine review

Project Structure

memory-firewall/
├── apps/
│   ├── api/
│   │   ├── app/
│   │   │   ├── main.py
│   │   │   ├── config.py
│   │   │   ├── deps.py
│   │   │   ├── routers/
│   │   │   │   ├── memories.py
│   │   │   │   ├── retrieval.py
│   │   │   │   ├── policies.py
│   │   │   │   ├── review.py
│   │   │   │   ├──

…

View on GitHub

Ways you can contribute:

• ⭐ Star the repository to support the project
• 🍴 Fork the project and experiment with new ideas
• 🐛 Open issues for bugs, improvements, or research discussions
• 🔧 Submit pull requests for features, fixes, or optimizations
• 💡 Share suggestions on memory governance, trust scoring, and memory security
• 📢 Share the project with others interested in AI agents and AI safety

Whether it's a small improvement or a major feature, every contribution helps move the project forward.