Skip to content

Query assistant

Choose a registered Neo4j database, then provide either a read-only Cypher query or a question in natural language.

  • Cypher query validates the statement with CyVer and executes it against the selected database.
  • Natural-language question searches all benchmark datasets associated with that database and returns the five semantically closest questions with their reference Cypher translations.

Read-only service

Only read Cypher queries are accepted and executions have a server-side timeout. The table shows the first 100 rows; download the full, untruncated result as JSON.

The KG valid result follows CyVer's definition: valid syntax, a schema score of 1, and a property score of 1 (or not applicable). Similarity percentages are cosine similarities computed with sentence-transformers/all-mpnet-base-v2 over the natural-language question field. You can pose a natural-language question, review the semantically similar ones from the benchmarks associated with the selected database, and then generate or refine the Cypher query with the help of an LLM.

For the natural-language mode, use the Examples to include in prompt slider (shown below the retrieved matches) to pick how many of them (1 to 5) to use as few-shot context, then Copy LLM prompt to copy a ready-made prompt to the clipboard. Paste it into an external LLM (e.g. ChatGPT) to obtain a Cypher query, then paste that query back here to validate and run it. Check Include Neo4j query output to run each retrieved example against the selected database and add its first three result rows to the prompt, giving the LLM concrete evidence of what each example query returns.