Elsai Graph Constructor

Package: elsai-graph-constructor  v0.1.0

Automatically builds knowledge graphs from unstructured text by identifying entities, discovering relationships, and formatting the data into a valid graph structure.

Installation

pip install --extra-index-url https://core-packages.elsai.ai/root/elsai-graph-constructor/ elsai-graph-constructor==0.1.0

Requirements: Python >= 3.10

---

GraphConstructor (high-level API)

The recommended entry point. Coordinates entity extraction, relationship discovery, and graph building in one call.

from elsai_graph_constructor import GraphConstructor

chunks = [
    {"chunk_id": "c1", "text": "John Doe works at Acme Corp in Paris."},
    {"chunk_id": "c2", "text": "Acme Corp is a leader in Aerospace technology."},
]

constructor = GraphConstructor(llm=llm, verbose=True)
success, message, graph_data = constructor.construct_graph(
    chunks=chunks,
    include_chunk_nodes=True,
)

if success:
    print(f"Nodes: {len(graph_data['nodes'])}")
    print(f"Edges: {len(graph_data['edges'])}")

llm can be any elsai-model connector, Bedrock, Gemini, or LangChain-compatible LLM.

---

Low-level components

EntityExtractor

Identifies key information points: people, organizations, locations, technologies, concepts.

from elsai_graph_constructor import EntityExtractor

extractor = EntityExtractor(llm=llm, verbose=True)
entities_by_chunk = extractor.extract_entities_batch(chunks)

# entities_by_chunk is a dict keyed by chunk_id
for chunk_id, entities in entities_by_chunk.items():
    print(f"{chunk_id}: {[e['label'] for e in entities]}")

RelationshipExtractor

Discovers connections between entities (e.g., Person WORKS_AT Company, Company LOCATED_IN City).

from elsai_graph_constructor import RelationshipExtractor

extractor = RelationshipExtractor(llm=llm, verbose=True)
relationships = extractor.extract_relationships_batch(
    chunks=chunks,
    entities_by_chunk=entities_by_chunk,
)

GraphSchemaBuilder

Formats extracted data into a valid graph structure, assigns IDs, and validates the result.

from elsai_graph_constructor import GraphSchemaBuilder

builder = GraphSchemaBuilder(verbose=True)

nodes = builder.build_nodes(entities_by_chunk, chunks=chunks)
edges = builder.build_edges(relationships, entities_by_chunk)

is_valid, message = builder.validate_graph(nodes, edges)
print(f"Graph valid: {is_valid} — {message}")

---

Output format

graph_data is a dictionary with:

{
    "nodes": [
        {"id": "John_Doe", "label": "John Doe", "type": "Person", "description": "..."},
        {"id": "Acme_Corp", "label": "Acme Corp", "type": "Organization", "description": "..."},
    ],
    "edges": [
        {"from": "John_Doe", "to": "Acme_Corp", "type": "WORKS_AT", "label": "Works At"},
    ]
}

Pass this directly to GraphStorage.store_graph() in elsai-graph-generator.