Vibe Coding Discoverwww-project-agent-memory-guard
View on GitHubOWASP Foundation web repository
OWASP Agent Memory Guard is a Python runtime defense layer that sits between an AI agent and its memory store, screening writes/reads for prompt injection, memory poisoning, secret leakage, protected-key tampering and self-reinforcement loops. It enforces YAML policies, emits audit events and ships drop-in integrations
Use Cases
Block memory poisoning in persistent AI agent memory storesDetect prompt injection arriving via tool outputs and memory writesRedact secrets and PII before they are persisted to agent memoryEnforce YAML policies (allow/redact/quarantine/block) over memory operationsGuard LangChain chat message history with drop-in middlewareProtect OpenAI Agents SDK, AutoGen, CrewAI and mem0 memory backendsDetect self-reinforcement poisoning loops in agent-authored writesEmit structured SecurityEvents for SIEM and audit correlationSnapshot and roll back memory to a known-good stateBenchmark agent memory defenses against attack payloadsProtect RAG knowledge stores from injection and tampering
Built With
- Language
- Python
- Frameworks
- LangChain · LangGraph · AutoGen · CrewAI · OpenAI Agents SDK · mem0 · LlamaIndex · FastAPI · Pydantic · PyYAML · uvicorn · Redis · OpenTelemetry · Transformers · PyTorch · pytest
Tags
agent-memory · ai-security · prompt-injection · guardrails · memory-poisoning · llm-security · owasp · ai-safety · agentic-ai · policy-enforcement · rag-security · middleware · forensics · mcp-server · opentelemetry · python