Find words in your knowledge and combine the matches with graph relationships.
Find relevant text in the workspace and use ranked matches as part of a graph query. Follow a match to its author, project, agreement, or other connected information.
Text search and graph traversal work together. You do not need to identify every resource by an exact identifier before exploring its connections.
Search for contracts mentioning flood damage, then restrict the results to policies for a particular building. The text finds the language; the graph supplies the relationship to the building.
The query's workspace and branch define its context. Access rules still apply to the knowledge returned.
Text matching is useful when particular terms matter. Similarity search finds related resources by a different measure. Choose the approach that answers the question, and inspect the results in their original context.
Read the query guide for the supported query syntax.
Developer reference
The index maps analyzed terms to postings for source literals/resources. Index-time and query-time Unicode normalization and tokenization must agree. Candidate IDs and relevance scores feed graph joins, allowing a lexical match to be constrained by RDF relationships and workspace authority.
Relevance is a ranking signal, not a truth probability. The index is a derived retrieval structure; source graph values remain authoritative.
Literal indexing, file-text extraction, named-graph coverage and historical-snapshot coverage are separate choices. Ingesting a PDF does not imply OCR, and retaining a commit does not imply an index for every historical version. A query must use the index coverage appropriate to its selected data state.