Glossary

Every term the lessons use is defined here in ordinary words. A lesson links here the first time it uses one of these, so nothing has to be explained twice and nothing is assumed.

Graph terms

Graph

A collection of items and the connections between them. A social network is a graph: the people are the items and the friendships are the connections.

Node

One item in a graph. A movie, a person, or a server is a node. Every node has a whole number called an ID that you use to refer to it later.

Edge

One connection between two nodes. An edge has a direction, so an ACTED_IN edge that points from a person to a film means something different from one pointing the other way.

Label

A category name attached to a node, such as Movie or Person. A node can carry more than one.

Property

A named detail stored on a node or an edge, such as a title or a year.

Traversal

Following edges from one node to the next in order to reach nodes further away.

Breadth-first search (BFS)

A traversal that explores a graph one level at a time: first everything one step from the start, then everything two steps away. Use it when you want the nearest things first.

Depth-first search (DFS)

A traversal that follows one path as far as it goes before backing up and trying another. Use it when you want to reach far parts of the graph quickly.

Shortest path

The fewest edges that connect two nodes.

Subgraph

A smaller piece cut out of a larger graph, usually everything within a few steps of some starting node.

Graph Query Language (GQL)

A written language for asking a graph database questions, in the same way that SQL asks questions of a table-shaped database.

Vector and search terms

Vector

A list of numbers.

Embedding

A vector that represents the meaning of something, such as a sentence or an image, as a point in space, arranged so that things with similar meanings sit close together. Comparing two embeddings is how a database judges whether two things mean roughly the same thing.

Dimension

How many numbers are in a vector. This site uses 768 throughout, because that is what the embeddinggemma model produces.

Cosine distance

One way of measuring how far apart two embeddings are. Smaller means more similar.

Vector search

Finding the stored items whose embeddings are closest to a query embedding, which is how you search by meaning rather than by keyword.

Hierarchical Navigable Small World (HNSW)

The structure AstraeaDB uses to find nearby embeddings quickly without comparing the query against every stored item.

Hybrid search

Ranking results by combining two signals: how close items are in the graph, and how similar their embeddings are.

Language-model terms

Large language model (LLM)

A program trained on a large amount of text that can write an answer to a question posed in ordinary language.

Retrieval-augmented generation (RAG)

Answering a question by first fetching relevant material and then giving it to a language model, so the answer is based on real source material rather than only on what the model absorbed in training.

GraphRAG

Retrieval-augmented generation where the fetched material is a subgraph rather than loose passages of text, so the model sees how facts relate to each other and every claim can be traced back to a node.

Linearization

Turning a subgraph into plain readable text so that a language model can be given it as input.

Practical terms

Server

The database program running in the background, holding your data.

Client

The library your own code uses to send requests to the server.

Port

A numbered channel on a machine that a server listens on. AstraeaDB uses 7687 by default, so the full local address is 127.0.0.1:7687, where 127.0.0.1 means this machine.

Write-ahead log (WAL)

A file the server appends to before changing its main data file, so that an interrupted write can be recovered rather than lost.