Upload documents and retrieve precise context for large language models (LLMs) through hybrid search—no servers to manage.
What is Knowledge Storage Service
Knowledge Storage Service is a fully managed Retrieval-Augmented Generation (RAG) service built on Tablestore. It provides an end-to-end pipeline—document import, parsing, chunking, vectorization, and hybrid search—that delivers precise context to LLMs.
The service uses a serverless architecture. Call APIs to create knowledge bases, upload documents, and run searches—no servers to deploy or maintain. Uploaded documents are automatically parsed, chunked, and indexed.
In a typical AI application, Knowledge Storage Service stores, indexes, and retrieves the knowledge that LLMs need.
Currently supported regions: China (Beijing), China (Shanghai), China (Hangzhou), China (Shenzhen), China (Hong Kong), and Singapore. Select the region closest to your workloads.
Use cases
Enterprise knowledge Q&A
Import product documentation, technical manuals, FAQs, and policies into a knowledge base, then pair it with an LLM to build an intelligent Q&A system. The system retrieves relevant chunks and generates accurate, evidence-based answers—applicable to customer service, HR, legal, IT operations, and other departments.
Intelligent document search and summarization
Replace keyword search with vector search that understands user intent and returns semantically relevant chunks. Combine with metadata filtering by time, category, author, and other dimensions for precise retrieval.
Multi-tenant SaaS knowledge base
Use the subspace mechanism to isolate data for different tenants (users, departments, or customers) within a single knowledge base. Each tenant searches only their own subspace—built-in data isolation without separate knowledge bases.
RAG pipeline integration
Use Knowledge Storage Service as the retrieval layer in a RAG architecture with frameworks such as LangChain and LlamaIndex. Import documents, call the Retrieve API during inference, and append the results to prompts before sending them to the LLM.
Compliance document management
Manage compliance documents for finance, healthcare, and government. All data stays in your own cloud account, meeting data residency requirements. Tag documents with metadata such as category, version, and expiration date, then filter by these attributes for precise retrieval.
Core strengths
Serverless and ready to use
Fully serverless. Create knowledge bases, upload documents, and run searches through API calls—no servers to provision or capacity to manage. A production-ready system takes a few API calls to set up.
Pay-as-you-go with zero minimum
No minimum specs or spending requirements. Storage and compute are billed independently by actual usage. Costs approach zero during low-traffic periods and scale automatically during peaks—no idle-resource charges.
Massive scale with no upper limit
A single knowledge base supports up to 100 million documents by default. Scale from small pilots to enterprise-grade repositories without capacity planning. Tablestore's distributed architecture provides horizontal scaling with no capacity bottlenecks.
Fully managed knowledge pipeline
Upload documents and the system handles parsing, intelligent chunking, embedding, and indexing automatically. Supported formats: PDF, Word (doc/docx), Excel (xls/xlsx), PowerPoint (ppt/pptx), plain text (txt), and Markdown (md). No need to build processing pipelines, deploy embedding models, or maintain vector indexes. The service manages the entire knowledge base lifecycle.
Full data ownership
All data—raw, intermediate, and result—is stored in your own cloud account (OSS bucket or Tablestore instance). The service never holds or replicates your data. You retain full ownership of your data. This architecture meets compliance requirements for regulated industries such as finance, government, and healthcare.
Flexible reranking for precise retrieval
Three built-in reranking strategies—Reciprocal Rank Fusion (RRF), weighted fusion, and model-based rerank—refine results across multiple dimensions and surface the most relevant content first. Hybrid search combines vector search (semantic similarity) with full-text search (exact keyword matching) to maximize retrieval coverage.
Open and customizable
All core modules are configurable through APIs and parameters. Control embedding model selection (built-in Model Studio or custom models), search strategy (type, retrieval count, and rerank method), metadata filtering, and chunk-level content editing and state management.
Comparison with self-managed solutions
|
Dimension |
Knowledge Storage Service |
Self-managed solution (open-source RAG) |
|
Deployment |
Serverless, ready to use through API calls |
Self-managed deployment with minimum hardware specs |
|
Dependencies |
None. OSS and Tablestore components are fully managed by Alibaba Cloud |
Self-managed Elasticsearch, MySQL, Redis, MinIO, and more |
|
Operations |
Zero maintenance with automatic capacity scaling |
Ongoing upgrades, monitoring, and failure recovery across multiple components |
|
Billing |
Pay-as-you-go with zero cost at zero usage |
Free software, but server, storage, and bandwidth costs apply |
|
Document processing |
Fully automatic (upload, parse, chunk, vectorize, and index) |
Manual parsing template and chunking strategy configuration |
|
Data security |
Data stored in your own OSS or Tablestore account |
Self-managed servers; security depends on operational practices |
|
Multi-tenancy |
Built-in subspace support, ready to use |
Requires custom isolation design |
|
Embedding |
Built-in Model Studio models (zero configuration) with custom model support |
Self-managed model API key setup |
|
Configuration complexity |
All parameters have optimized defaults, ready to use with zero configuration |
Many parameters to learn and tune; high onboarding cost |