Show HN: Membrane, revisable memory for long lived AI agents (github.com)

Membrane, a newly introduced memory substrate for long-lived AI agents, is designed to overcome the limitations of traditional agent memory systems, which often rely on ephemeral context windows or static logs. Unlike these approaches, Membrane enables selective and revisable memory, allowing AI agents to capture raw experiences and transform them into structured knowledge. Its architecture supports the revision of facts and procedures with evidence, promoting continuous learning, safety, and predictability for AI systems. This development is significant for the AI/ML community as it provides a framework that not only retains knowledge but also ensures that it remains accurate and relevant over time. Key technical features include the ability to consolidate episodic traces into various memory types—episodic, semantic, competence, and plan graphs—while implementing mechanisms like decay salience, trust-aware retrieval, and sensitive data handling with encryption. Membrane's gRPC API facilitates seamless integration and interaction with existing systems, making it a robust tool for building more resilient and adaptable AI agents capable of ongoing improvement.