Memlineage v0.1.0: Two-Layer Defense Against LLM Memory Poisoning

Memlineage v0.1.0 is a Python library that defends LLM agents against memory poisoning attacks using cryptographic provenance and behavioral detection. It uses Ed25519 signatures to tie memories to their origin and a graph-based trust propagation that prevents untrusted memories from ever becoming trusted. An independent audit confirmed the system blocks spoofing, transitive laundering, and mixed justifications.