Knowledge and Memory Storage is a Serverless data service that Tablestore provides for AI agents. It includes knowledge storage and memory storage, supplying retrievable knowledge for RAG applications and helping agents maintain context across sessions.
What is Knowledge and Memory Storage
Knowledge and Memory Storage is a Serverless data service that Tablestore provides for AI agents. The knowledge storage service transforms business documents into retrievable knowledge to supply context for RAG applications. The memory storage service stores and recalls user preferences, historical facts, and session information, helping agents maintain context across sessions.
You can use the two services independently or in combination. For example, an intelligent customer service bot can store product manuals and FAQs in a knowledge base while storing user preferences and interaction history in a memory store, so that answers draw on both authoritative materials and personalized context at the same time.
Product services
Depending on your business needs, you can choose the knowledge storage service, the memory storage service, or a combination of both.
|
Scenario |
Choice |
Typical data |
Core workflow |
|
Let an agent remember preferences, facts, experiences, and session context, or organize materials into an Agent Wiki (knowledge graph) |
Conversation messages, long-term memory, memory files, Wiki pages |
Write messages, files, or Wiki sources → generate, organize, and recall by Scope |
|
|
Answer questions based on product manuals, policies, contracts, and other materials |
Documents such as PDF, Word, Markdown, and images |
Create a knowledge base → import documents → wait for indexing → retrieve chunks |
|
|
Obtain both reliable materials and personalized context |
Both combined |
Enterprise documents + user interactions |
Retrieve knowledge first, then recall the relevant memories of the current user |
A knowledge base stores authoritative materials that are maintained by the business and relatively stable. A memory store stores context that changes continuously with interactions and must be isolated by application, tenant, agent, or session.
Core capabilities
-
Serverless fully managed
No need to deploy or maintain underlying storage, document processing, embedding, or retrieval components. You can use knowledge and memory capabilities through the console, SDK, or command-line tool. Underlying resources scale elastically with your business volume.
-
Automated processing and hybrid retrieval
The knowledge storage service can automatically perform document parsing, chunking, embedding, and index building. It supports vector retrieval, full-text retrieval, hybrid retrieval, and multiple reranking strategies.
-
Multi-modal memory management
The memory storage service supports short-term messages, long-term structured memory, file memory, memory consolidation (Dream), and knowledge compilation (Wiki). You can save, organize, and recall different forms of context based on your business needs.
-
Multi-tenant data isolation
The knowledge storage service isolates data of different tenants or business domains through Subspaces. The memory storage service isolates application, tenant, agent, and session data through a four-level Scope.
Billing
Knowledge and Memory Storage uses a pay-as-you-go model. Charges include data storage fees and Credit fees generated by operations such as document embedding, retrieval reranking, and memory processing. For details, see Product Billing.
Get started
For first-time use, you can choose either of the following paths to complete a minimal workflow:
For information about instances, networking, and access control, see AgentStorage instances. For identity authentication and permission configuration, see API Key management and Configure access permissions.
FAQ
What is the difference between knowledge storage and memory storage?
Knowledge storage primarily manages relatively stable business materials such as product manuals, policies, and contracts, and provides retrieval results for RAG applications. Memory storage primarily manages preferences, facts, and session context that change continuously with user interactions.
Do I have to use knowledge storage and memory storage at the same time?
You can choose flexibly based on your business scenario. There is no product-level restriction. You can use either service independently based on your business needs, and combine both when you need authoritative materials and personalized context at the same time.
Do I need to deploy a vector database myself?
No. Knowledge and Memory Storage provides underlying storage, indexing, and retrieval capabilities through Tablestore. You do not need to deploy or maintain a vector database yourself.
Can I use an existing Tablestore instance?
Yes. When you initialize a client through the SDK or command-line tool, you can specify an existing Tablestore instance to store knowledge base or memory store data. If you do not specify one, the system automatically creates a service-managed Tablestore instance. The two options are billed differently. For details, see Product Billing.
Which regions are currently supported?
Knowledge and Memory Storage currently supports the China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Hong Kong), and Singapore regions.