Knowledge Graph
Ask the configured LLM to extract a knowledge graph — entities and the relations between them — from the input. Useful for building a queryable representation of unstructured documents.
Pre-requisite: Install an AI provider application from Profile > {Organization} > Applications.
Parameters
Model identifier from the selected provider. Defaults to the provider's recommended model when left empty.
System prompt prepended to every call. Use it to set extraction rules (e.g. relation vocabulary, entity types).
JSON path pointing to the array to iterate over (e.g. $.pages[*]). The
path is queried against the input and each matched item is processed
independently into nodes and edges, with the graph built up incrementally
across items. A path that is invalid or matches nothing fails the step
with a JSONPath error — it does not fall back to feeding the whole input
to the model.
Domain-specific knowledge sent alongside each call (e.g. "These are pharmaceutical clinical trial reports — entities of interest are drugs, conditions, dosages, outcomes."). Helps the model produce a relevant graph.
Input
Source content — either a JSONL file or an inline JSON record.
Output
JSON document of the form {"nodes": [...], "edges": [...]}. Each node
carries id, label, node_type, properties, source_citations
(snippet, character range, and the page_number of the matched item it
came from), and a confidence_score between 0 and 1. Each edge carries
source, target, relationship, properties, an optional
source_citation, and a confidence_score.