MiniIO AIStor Memory targets persistent storage for AI agents
MiniIO launched AIStor Memory to keep agent context on customer-controlled infrastructure, but pricing and benchmarks were not disclosed.
By Dominic Okoye · Staff Writer
· 3 min read
MiniIO AIStor Memory is the object storage company’s new product for giving AI agents persistent context across sessions, a problem that gets more expensive as enterprises move from chatbots to longer-running agent workflows. MiniIO says the product keeps agent-generated work on infrastructure controlled by the customer, an important point for companies trying to govern regulated or sensitive data.
The company is positioning AIStor Memory as a foundation layer for agentic AI systems, where agents may create documents, analyze information, answer complex questions or work through multistep tasks that continue over hours or days. MiniIO argues that those outputs become part of a company’s operating knowledge, and that enterprises need to control where that memory lives and how it is secured.
What is MiniIO AIStor Memory?
AIStor Memory is designed to store agent memory as a native data type alongside objects and tables, according to MiniIO. That means an agent can retain work state across sessions and resume an unfinished task while using existing enterprise controls rather than relying only on the short-term context passed into a model during inference.
MiniIO co-founder and co-Chief Executive AB Periasamy said in the company’s announcement that AI-generated knowledge should remain on enterprise-controlled infrastructure. He said AIStor Memory combines long-term memory, persistent workspaces and secrets on one customer-controlled foundation, and described a single agent’s memory as shared infrastructure for the broader organization.
The product addresses a messy systems problem. MiniIO says companies trying to give agents durable memory often stitch together object storage, metadata databases, vector stores, secrets managers, governance tools and synchronization pipelines. AIStor Memory is pitched as an integrated option that can be mounted onto existing sandboxes and used with current AI infrastructure, tools and frameworks without modifications, according to the company.
Why agent memory is becoming an infrastructure issue
Conventional chatbots can often operate inside a narrower context because the interaction is usually a prompt-and-response exchange. Agentic systems are different: they may need to keep intermediate files, decisions, credentials and task history available across interruptions, human approvals and multiple sessions.
MiniIO says AIStor Memory is built for use cases such as software engineering agents working across large codebases, research workflows that take days, human-in-the-loop processes that must pause and restart, and enterprise AI systems handling regulated information. The company claims the product sits below the KV-cache layer served by its MemKV service, preserving the full state of an agent’s work rather than only the context used during inference.
That architecture is how MiniIO says it can support what it calls “infinite context,” with memory limited by available storage capacity rather than a model’s context window. The company says the approach avoids summarizing, truncating or otherwise reducing the stored work state.
MiniIO also said AIStor Memory uses erasure coding, bitrot protection, encryption, compression and fault tolerance to protect data against rack, drive and data center failures. Because the system runs on customer infrastructure and uses customer-owned keys, MiniIO says organizations can reduce concerns about agent memory leaking outside their control. The company did not disclose pricing, customer adoption figures or performance benchmarks for the product.
This story draws on original reporting from SiliconANGLE.
