Vector Database

A vector database is a data store built to index embeddings and answer nearest-neighbor queries: given a query vector, return the stored vectors closest to it by a measure such as cosine similarity.

Why it matters

Searching millions of high-dimensional vectors by brute force is slow. Vector databases use approximate nearest neighbor indexes (for example HNSW) to trade a little accuracy for large speed gains. They are a common building block of RAG, recommendations, and image search.

Example

A typical flow:

1. Split documents into chunks
2. Embed each chunk -> store vector + text + metadata
3. Embed the user question
4. Query: top 5 nearest chunks
5. Pass those chunks to the language model

Common misconceptions

• You always need a dedicated product. Some relational databases offer vector extensions that suit small to medium workloads.
• Results are exact. Approximate indexes can miss true nearest neighbors.
• It replaces keyword search. Many systems combine both for better results.

Related

• Embedding
• Retrieval-augmented generation
• Metadata
• Latency

यह पेज सामान्य जानकारी है, जिसे हमारी संपादकीय टीम ने लिखा है और दस्तावेज़ीकरण बदलने पर अपडेट किया जाता है। सर्वर और प्रदाता अलग-अलग होते हैं, इसलिए बदलाव पहले सुरक्षित तरीके से जाँचें। देखें: संपादकीय नीति और अस्वीकरण.