MinIO has launched AIStor Memory, a feature for its object storage software aimed at giving AI agents persistent memory so they can resume multi-step tasks across sessions rather than losing context between them. The company argues that as AI agents take on work such as decision-making, drafting documents and analysis, the output they generate constitutes valuable "organisational knowledge" that businesses need to retain and control on their own infrastructure rather than leaving it scattered across third-party tools.

AIStor Memory treats agent memory as a native data type alongside objects and tables, combining long-term memory, persistent workspaces and secrets management in one system that mounts into existing sandboxes without requiring changes to current tools or frameworks. MinIO says this replaces the usual patchwork of vector stores, metadata databases, secrets managers and synchronisation pipelines, and highlights use cases including software engineering agents working across large codebases, multi-day research tasks, human-in-the-loop workflows, and regulated enterprise data handling. The system offers "infinite" context that scales with storage capacity rather than a model's context window, standard HTTPS or POSIX access, and enterprise-grade durability via erasure coding, encryption and fault tolerance, with data kept under customer-held keys.