Vector Database
A vector database is a database system specialized for storing and retrieving high-dimensional vector data. It persistently stores embedding representations of unstructured data such as text, images, and audio, and supports approximate nearest neighbor (ANN) search based on similarity metrics such as cosine similarity and Euclidean distance. Vector databases are widely applied in semantic search, recommendation systems, image retrieval, and retrieval-augmented generation for large language models.
Core Facts
RAG（检索增强生成）技术架构将文档切片、向量化、存入向量数据库，查询时先检索相关片段再交给模型生成回答
An Agent's memory mechanism is typically divided into two categories: short-term memory and long-term memory, with long-term memory requiring external persistent storage such as vector databases, relational databases, and graph databases.
Vector databases such as FAISS, Chroma, and Weaviate build indexes by encoding text as high-dimensional vectors, enabling semantic similarity retrieval in milliseconds.
Timeline (last 90 days)
Current mainstream RAG implementations mostly use vector databases to perform dense retrieval over unstructured text, while their retrieval support for structured data sources such as tables and relational databases is relatively weak.
Interviewers may ask questions such as 'What is a vector database?', 'How does RAG mount a knowledge base?', and 'What is MCP?', which have become must-ask questions for large-model-oriented positions.
A vector database is a database specifically for storing and retrieving high-dimensional vector data; in AI applications, unstructured data such as text and images is converted into vectors by embedding models and stored, and at query time the closest content is found by computing cosine similarity.
Hillock is not positioned to replace vector databases, but to introduce deterministic judgment at the retrieval layer.
Compared with traditional relational databases, vector databases have higher storage and query overheads, and are usually billed by vector dimension, index scale and number of queries.
The core capability of vector databases is approximate nearest neighbor search (ANN); Milvus supports billion-scale vectors, and PGVector suits teams already using PostgreSQL.
The content stored in vector databases is completely unreadable to users, who cannot directly view, edit entry by entry, or delete memory items.
Mainstream vector database solutions do not differ much in small and medium-scale scenarios, and getting the ingestion, retrieval, and reranking pipeline working is more valuable than agonizing over selection.
Common practices for isolating long-term memory by user include metadata filtering by user ID in a vector database, creating a separate collection for each user, or enforcing tenant constraints at the retrieval layer.
Context engineering is essentially information scheduling within a limited window: deciding what to keep, what to compress, and what to move out of the window and recall on demand from external storage (such as a vector database).
40 more timeline events
All Facts (20)
RAG（检索增强生成）技术架构将文档切片、向量化、存入向量数据库，查询时先检索相关片段再交给模型生成回答
90%VerifiedAn Agent's memory mechanism is typically divided into two categories: short-term memory and long-term memory, with long-term memory requiring external persistent storage such as vector databases, relational databases, and graph databases.
80%VerifiedVector databases such as FAISS, Chroma, and Weaviate build indexes by encoding text as high-dimensional vectors, enabling semantic similarity retrieval in milliseconds.
80%VerifiedRetrieval-Augmented Generation (RAG) splits codebases into semantic segments via vector databases and dynamically retrieves the most relevant content to inject into the context when an AI handles specific tasks.
80%VerifiedVector memory databases convert historical conversation content into high-dimensional semantic vectors, and during new conversations dynamically retrieve relevant memory fragments via similarity search and inject them into the context window.
80%VerifiedRAG addresses the limitation of large models' knowledge cutoff dates by splitting private documents into text chunks, converting them into vector representations stored in a vector database, and injecting retrieved relevant content into the Prompt.
80%VerifiedRAG（检索增强生成）的核心思想是在大模型生成回答之前，先从外部知识库中检索相关文档片段作为上下文注入Prompt
80%VerifiedAgent的记忆系统通常分为三个层次：工作记忆（当前对话上下文窗口）、短期记忆（近期交互历史）和长期记忆（通过向量数据库持久化存储）
80%VerifiedVector databases store high-dimensional vectors converted from text by an Embedding model, typically 768 or 1536 dimensions.
80%VerifiedVector databases (such as ChromaDB, Milvus, and Pinecone) use indexing algorithms such as HNSW and IVF to achieve millisecond-level retrieval among millions or even hundreds of millions of vectors.
80%Verified常见的向量数据库包括ChromaDB、Milvus、Pinecone、Weaviate，它们针对高维向量的近似最近邻（ANN）搜索进行了专门优化
80%VerifiedThe pgvector extension adds vector storage and similarity search capabilities to PostgreSQL, enabling Supabase to serve simultaneously as a traditional database and a vector database.
80%VerifiedVector databases (such as Pinecone, Weaviate, Milvus, etc.) convert text into high-dimensional vectors via embedding models and use cosine similarity or Euclidean distance for approximate nearest neighbor (ANN) search.
80%VerifiedEmbedding是将文本转换为高维数值向量的过程，语义相近的文本在向量空间中距离更近
80%Verified长期记忆Agent通常基于RAG（检索增强生成）技术构建，通过向量数据库将信息转化为高维向量并持久化存储
80%Verified向量数据库的核心操作是近似最近邻搜索（ANN），通过HNSW、IVF等索引算法将相似度计算加速到毫秒级响应
80%Verified向量数据库将文本内容通过嵌入模型（Embedding Model）转化为高维数值向量，检索时通过余弦相似度等算法找出语义最接近的内容
80%VerifiedAI memory implementation approaches are broadly divided into two categories: local device memory and cloud server memory; local solutions offer better privacy but are limited by device compute power and difficulties in cross-device synchronization, while cloud solutions are more capable but carry higher risks of user data exposure.
75%VerifiedRAG (Retrieval-Augmented Generation) introduces external memory during the inference phase to constrain the model's generation boundaries, by chunking documents, embedding them as vectors stored in a vector database, and retrieving relevant chunks to inject into the prompt at inference time.
75%VerifiedVector retrieval relies on vector databases for fuzzy matching and cannot precisely express structural relationships, whereas graph retrieval uses graph databases as the storage medium and is specifically optimized for multi-hop traversal queries.
75%