Commonplace: Self-hosted, privacy-tiered memory for your AI agents (github.com)

Commonplace has announced a self-hosted, privacy-focused memory system for AI agents, utilizing a two-tier Graphiti knowledge graph accessible via a private Tailscale network. This solution operates entirely on local hardware, requiring only a single Linux host equipped with Docker and a consumer NVIDIA GPU. The system is offline-first, ensuring that sensitive data, including entity extraction performed by a language model (LLM), remains secure within the user’s local environment. It segregates memory into two tiers: the personal tier for general use and a client-confidential tier for sensitive information, both leveraging the same embedding model to optimize performance while maintaining data isolation. This development is significant for the AI/ML community as it emphasizes the importance of privacy and data security in AI applications. By incorporating an offline architecture and enabling data processing to occur entirely on local hardware, Commonplace mitigates risks associated with cloud computing, making it especially appealing to enterprises handling confidential information. The architecture employs a slow extraction method that does not impact query latency, maintaining efficiency while ensuring that sensitive operations are performed securely and privately.