Milvus

Milvus — explained in detail

Milvus is an open-source vector database — a database system built specifically to store embeddings (numeric vectors) and to find, very quickly, the vectors most similar to a query vector. This similarity search over high-dimensional vectors underpins many AI applications, in particular retrieval-augmented generation (RAG) and semantic search.

The project was originally started by the company Zilliz and is today part of the LF AI & Data Foundation (under the Linux Foundation umbrella). The source code is released under the permissive Apache 2.0 license; Zilliz remains the main contributor and also offers a managed variant called Zilliz Cloud.

Architecture and core function

Milvus is built to be distributed and cloud-native: storage and compute are separated, so the system can scale horizontally across multiple nodes — from a local single instance up to clusters with billions of vectors. At its core is approximate nearest neighbor (ANN) search: instead of comparing the query vector exactly against every stored vector (too slow at large scale), Milvus uses index structures that find the nearest neighbors approximately but very fast.

Milvus supports several index types with different trade-offs between speed, memory and accuracy — including HNSW (low latency), IVF variants (compressed/exact), DiskANN (disk-based for large datasets) and GPU-accelerated indexes. It also offers metadata filtering, hybrid search and scalar fields alongside the vectors.

Context

Milvus is one of the most widely used dedicated vector databases and is often cited as an open-source alternative to managed services such as Pinecone. It is actively developed; a Milvus 3.0 generation was in progress in mid-2026.

Example / Practical use

In a RAG application Milvus typically works like this: documents are split into chunks, each chunk is converted into a vector by an embedding model and stored in Milvus. When a user asks a question, the question is also turned into a vector; Milvus returns the most similar chunks (measured e.g. via cosine similarity), which are then passed to the language model as context. Metadata filters can narrow the search further, for example to a specific source or language.

Distinction from related terms

• Milvus vs. vector database (general): A vector database is the product category; Milvus is one concrete, open-source product within it.
• Milvus vs. Pinecone: Pinecone is a purely managed cloud service (closed source); Milvus is open source and self-hostable — with Zilliz Cloud as an optional managed variant.
• Milvus vs. Qdrant / Weaviate / Chroma: All are dedicated vector databases; they differ in architecture, scaling model, supported indexes and maturity. Milvus is particularly geared toward very large, distributed deployments.
• Milvus vs. pgvector: pgvector is an extension for the relational database PostgreSQL, not a standalone system. Sufficient for smaller datasets; Milvus specialises in pure, highly scalable vector workloads.

Entdecke mehr

Blog

Saving Tokens with Claude: 6 Principles That Make Experts Twice as Fast

How I turned my CLAUDE.md from a style guide into a token budget — 6 principles for lower cost, less waiting, and more honest reporting.

Glossar

Chroma

Chroma (ChromaDB) is an open-source vector database for storing and searching embeddings. It is widely used in RAG systems to retrieve relevant text passages via similarity search and feed them to language models.

Lexikon

Vector Databases Compared

What vector databases do, when you need one, and how Chroma, Weaviate, Milvus, Qdrant and pgvector stack up against each other.

Themenuebersicht

KI-WerkzeugeVektor-Datenbanken