Tutorial: Building with CleaveDB

Graph & Semantic Search (FIND) — Overview

Some questions are about the contents of one bucket; others are about how documents connect or what an unfamiliar phrase means. CleaveQL’s FIND command is designed for that second kind of exploration. It follows bonds through CleaveDB’s document graph and searches text by semantic similarity, so an application can ask about paths and concepts rather than only exact field values.

Follow a connection

CleaveDB stores relationships as directed bonds between documents. A direct lookup follows one named bond from a known source; a graph pattern describes several linked steps and any conditions that the resulting path must satisfy. These forms are useful in social graphs, ownership chains, product provenance, and other data where the connection is part of the answer.

Search for an idea

Semantic FIND begins with a phrase and searches documents whose text has related meaning, even when the stored wording differs. CleaveDB creates document embeddings in the background and compares the query embedding with them. This is a natural fit for discovery and natural-language search, with the practical detail that a freshly written document may take time to appear in semantic results.

Explore the FIND lessons

Each topic has its own page with CleaveQL examples, an explanation of how the query works, and guidance for using it well:

• Following a bond: retrieve a document connected to a known source.
• Graph patterns: trace multi-step paths and constrain the matching nodes.
• Semantic search: retrieve documents by similarity of meaning.
• Historical graph reads: inspect a relationship as it existed earlier.
• Performance and edge cases: understand graph growth, indexing, and vector-search timing.