Show HN: Kremis – Graph-based memory for AI agents with no hidden state (Rust) (github.com)

Kremis has been introduced as a cutting-edge graph-based cognitive substrate built in Rust, designed to enhance the way AI agents manage memory without relying on hidden states. Unlike traditional AI systems that may fabricate data or operate as black boxes, Kremis ensures that every output is traceable to actual ingested signals, offering a deterministic behavior with no randomness. This explicit structure allows for full inspectability, addressing issues like hallucinations, opacity, and grounding, making it a promising tool for developers focused on reliable and accountable AI systems. This innovative system is significant for the AI/ML community as it eliminates common pitfalls associated with conventional AI architectures. By implementing ACID transactions and a minimalistic design that only retains structural relationships from real-world data, Kremis positions itself as a trustworthy framework for creating applications where understanding and interpretation of data are critical. Additionally, Kremis can be integrated with AI assistants, such as Claude, through a Model Context Protocol (MCP) server, further extending its utility in advanced AI applications. Its focus on being crash-safe while maintaining a compact size enhances its appeal for developers aiming to build robust AI solutions.