Add agent-memory-guard

[vgudur-dev]

Category

Security > Web Security

Project

agent-memory-guard - Runtime security scanning for AI agent memory stores detecting injection, poisoning, and leakage.

Checklist

• One project per PR
• PR title format: Add project-name
• Entry format: - [project-name](url) - Description ending with period.
• Description is concise and short

Why This Project Is Awesome

Which criterion does it meet? (pick one)

• Industry Standard - The go-to tool for a specific use case
• Rising Star - 5000+ stars in < 2 years, significant adoption
• Hidden Gem - Exceptional quality, solves niche problems elegantly

Explain:

Agent Memory Guard is the only runtime security scanner specifically designed for AI agent memory stores. It's an official OWASP Incubator Project (the reference implementation for OWASP ASI06 — Memory Poisoning in the OWASP Top 10 for Agentic AI).

Why it qualifies as a Hidden Gem:

1. Solves a unique, critical problem — As AI agents gain persistent memory (LangChain, AutoGen, CrewAI, mem0), there's no other tool that scans memory writes for prompt injection, data leakage, privilege escalation, and poisoning at runtime. This is the first and only library in this space.

2. Production-ready quality — v0.3.0 ships with CLI scanner (amg scan), REST API server, ML-based detection, 6 detector categories, YAML policy engine, and LangChain middleware. 92.5% detection rate, 0% false positives, 59µs median latency.

3. Real-world usage — 4,100+ PyPI downloads, LangChain middleware integration published, GitHub Action for CI/CD scanning, used in production by teams building agentic AI systems.

4. OWASP backing — Official OWASP project with structured governance, not a weekend project. Repository is 4+ months old with 155 commits and consistent activity.

5. Zero dependencies — Pure Python, no API keys, runs locally. Drop-in integration with any framework via the MemoryStore protocol.

How It Differs

No similar entry exists in awesome-python. The closest entries are in the "AI and Agents > Data Layer" section (mem0, llama-index, instructor) — but those are memory providers. Agent Memory Guard is a security scanner that wraps any memory provider to detect attacks. It's complementary, not competitive.

[JinyangWang27]

@vgudur-dev Thanks for your submission. However, this project does not meet our acceptance criteria:

Repository age: The repository was created 4 months ago. We require a minimum of 6 months for Hidden Gem submissions to ensure project stability.

GitHub stars: 34 stars (minimum 100 required, or strong justification for Hidden Gem).

Please see our CONTRIBUTING.md for full requirements. You're welcome to resubmit once the project has matured and gained community traction.