New to vector databases? Vector databases store numerical representations of your data called embeddings, and find results by similarity rather than exact keyword matching. Use one when meaning matters more than exact text: semantic search, RAG pipelines, agent memory.
Architecture
VectorAI DB runs as a single Docker container with no external dependencies. Data is persisted to a volume-mounted directory on the host.| Port | Protocol | Purpose |
|---|---|---|
| 6573 | HTTP | REST API |
| 6574 | gRPC | gRPC API (recommended for SDKs) |
| 6575 | HTTP | Local UI |
Data model
| Concept | What it is |
|---|---|
| Collections | Named containers for vectors, similar to tables in a relational database. Each collection has a fixed vector dimension and distance metric set at creation time. |
| Points | The unit of data in a collection. Each point has a unique ID, a vector embedding, and an optional JSON payload. |
| Vectors | Fixed-length arrays of floats that represent your data semantically. Generated by embedding models from text, images, or other content. |
| Payload | JSON metadata attached to a point. Use payloads to store context and to filter search results. |
| Distance metrics | The function used to measure similarity between vectors. Supported metrics: Cosine, Euclidean, and Dot Product. |
How a search works
Generate an embedding
Your application converts a query (text, image, or other data) into a vector using an embedding model.
HNSW nearest-neighbour search
VectorAI DB traverses the HNSW index to find the closest vectors, applying any payload filters inline.
- Semantic search for meaning-based retrieval
- Hybrid search to combine dense and sparse scores
- Filtering for must, should, and must-not payload conditions
Next steps
Installation
Install VectorAI DB with Docker.
Quickstart
Create a collection, insert vectors, and run a search.
Fundamentals
Collections, points, vectors, search, and filtering.
Academy
Hands-on tutorials for semantic search, RAG, and more.