AIOZ Storage Agent Memory: Record and Recall Context by Meaning

AIOZ Storage Agent Memory: Record and Recall Context by Meaning

AIOZ Storage Agent Memory is now LIVE.

Create a memory bucket and give agents a way to retain and retrieve relevant context across sessions and tasks.

Record: stores a statement verbatim and completes the write synchronously.

Recall: retrieves relevant memories by semantic meaning and returns a similarity score with each result.

Persistent Memory for Agent Workflows

Agent workflows need useful context to remain available beyond a single session or context window.

AIOZ Storage Agent Memory provides a purpose-built layer for storing context and recalling it later by meaning.

The core workflow stays simple:

  1. Create an AIOZ Storage Agent Memory bucket.
  2. Record statements the agent may need later.
  3. Recall relevant memories with a natural-language question.

The returned memories preserve the original stored text and include similarity scores for each match.

What You Can Build

AIOZ Storage Agent Memory supports applications that need context to remain available across sessions, tasks, and coordinated workflows.

  • Cross-Session Assistants: Keep useful information available after a conversation ends.
  • Preference Memory: Preserve recorded user preferences for later recall.
  • Multi-Agent Context: Retrieve decisions and contextual notes across coordinated agent workflows.
  • Customer Support Memory: Keep relevant customer context available for later interactions.
  • Long-Running Workflows: Carry useful context beyond a single context window.

Built on AIOZ Storage Vector Database

AIOZ Storage Agent Memory is built on top of AIOZ Storage Vector Database, the general-purpose storage and retrieval layer for vector embeddings.

Agent Memory adds a focused interface for agent workflows by fixing the embedding model and removing the need to choose a vector dimension, configure a distance metric, or build an index before storing memories.

Every Agent Memory bucket uses multilingual-e5-small, enabling a question in one language to match a memory recorded in another without separate model selection.

Built Around Familiar Storage Tools

Statements and recall questions can each contain up to 32 KiB of text.

Memory buckets accept only the two memory operations, record() and recall(), rather than arbitrary raw-vector writes. Access is enforced at the bucket level.

Bucket management uses AWS Signature Version 4 signing and standard AWS SDK compatibility. Dedicated SDKs are also available for Go, JavaScript / TypeScript, and Python from a shared API specification.

Pay-As-You-Go Agent Memory

Current listed pricing includes:

  • Vector storage: $0.05 per GB-month, charged hourly
  • Vector upload: $0.04 per GB uploaded
  • Query requests: $0.000002 per query
  • Data returned: $0.01 per GB returned, with the first 512 KB per query free

There is no monthly plan required. Memories remain available until the memory bucket is deleted, with no storage epoch or renewal cycle.

Start with a Memory Bucket

Create a memory bucket. Record what matters. Recall it by meaning.

Persistent memory, built into AIOZ Storage.

Build agents that remember with AIOZ Storage Agent Memory.