ollama - 💡(How to fix) Fix [Feature Request] Memory Poisoning Protection for Ollama Agents via OWASP Agent Memory Guard

Problem

Ollama-powered agents with persistent memory (via LangChain, LlamaIndex, or custom pipelines) are vulnerable to memory poisoning attacks. Adversarial inputs stored in conversation history or vector stores can cause local LLM agents to leak system prompts, ignore safety instructions, or produce corrupted outputs.

This is especially critical for Ollama users running agents locally where there's no cloud provider safety net.

Proposed Solution

OWASP Agent Memory Guard (AMG) is an open-source Python library that wraps any memory store as a transparent security layer:

• pip install agent-memory-guard
• Scans every memory write for prompt injection, PII leakage, and tampering
• 92.5% detection rate on AgentThreatBench
• Works with Ollama + LangChain, LlamaIndex, or any custom pipeline
• Runs entirely locally — no cloud dependency

Links

• GitHub: https://github.com/OWASP/www-project-agent-memory-guard
• PyPI: https://pypi.org/project/agent-memory-guard/
• Benchmark: https://pypi.org/project/agent-threat-bench/