Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.clarifeye.ai/llms.txt

Use this file to discover all available pages before exploring further.

​
Search Operations

Perform semantic search and Cypher queries against your warehouse data. Perform semantic search across warehouse data: 
warehouse = client.get_warehouse("warehouse-id")

response = warehouse.retrieve_with_semantic_search(
    query="financial reports from Q4",
    n_objects=10,
    indexes=["chunks"],  # Optional, defaults to ["chunks"]
    reformulate_query=True, # Optional, defaults to False
    enrich_with_chunks=True, # Optional, defaults to False
    included_tags = [{"parent": "job-category", "children": "data-engineer"}],
    excluded_tags = [{"parent": "job-category", "children": "data-engineer"}],
    rerank=True, # Optional, defaults to False
)

results = response.get("retrieval_results", [])
for result in results:
    print(f"Score: {result.get('score', 'N/A')}")
    print(f"Content: {result.get('content', '')[:100]}...")
Parameters:
  • query (str): Natural language search query
  • n_objects (int, optional): Number of results. Default: 10
  • indexes (list[str], optional): Specific indexes. Default: [“chunks”]
  • reformulate_query (bool, optional): Whether to perform HyDE query reformulation before retrieving objects. Default: False
  • enrich_with_chunks (bool, optional): Whether to enrich the retrieved objects with their original chunks. Default: False
  • included_tags (list[dict], optional): Tags to include in the search. Default: []
  • excluded_tags (list[dict], optional): Tags to exclude in the search. Default: []
  • rerank (bool, optional): Whether to perform reranking on retrieved objects. Default: False
  • included_objects (list[dict], optional): Objects to include in the retrieval. A list of [{"name": entity_name}]. Default: []
  • excluded_objects (list[dict], optional): Objects to exclude in the retrieval. A list of [{"name": entity_name}]. Default: []
  • selected_documents (list[dict], optional): List of {"id": document_uuid} to select from. Default []
Details on the Tags (included or excluded) There are 2 possibilities for the identification of tags:
  • either by a tag id and its parent, for example: {"parent": "job-category", "children": "data-engineer"}
  • or by its hierarchy and its id, for example {"id": "fullstack-engineer","id_tree": "job-roles"}
Returns: A list of retrieved objects.

​
Full text search

Performs a full text search on the knowledge graph fulltext indexes. 
warehouse = client.get_warehouse("warehouse-id")

response = warehouse.retrieve_with_full_text_search(
    query="financial reports from Q4",
    n_objects=10,
    indexes=["chunks"],  # Optional, defaults to ["chunks"]
    reformulate_query=True, # Optional, defaults to False
    enrich_with_chunks=True, # Optional, defaults to False
    included_tags = [{"parent": "job-category", "children": "data-engineer"}],
    excluded_tags = [{"parent": "job-category", "children": "data-engineer"}],
    rerank=True, # Optional, defaults to False
)

results = response.get("retrieval_results", [])
for result in results:
    print(f"Score: {result.get('score', 'N/A')}")
    print(f"Content: {result.get('content', '')[:100]}...")
Parameters:
  • query (str): Natural language search query
  • n_objects (int, optional): Number of results. Default: 10
  • indexes (list[str], optional): Specific indexes. Default: [“chunks”]
  • reformulate_query (bool, optional): Whether to perform HyDE query reformulation before retrieving objects. Default: False
  • enrich_with_chunks (bool, optional): Whether to enrich the retrieved objects with their original chunks. Default: False
  • included_tags (list[dict], optional): Tags to include in the search. Default: []
  • excluded_tags (list[dict], optional): Tags to exclude in the search. Default: []
  • rerank (bool, optional): Whether to perform reranking on retrieved objects. Default: False
  • included_objects (list[dict], optional): Objects to include in the retrieval. A list of [{"name": entity_name}]. Default: []
  • excluded_objects (list[dict], optional): Objects to exclude in the retrieval. A list of [{"name": entity_name}]. Default: []
  • selected_documents (list[dict], optional): List of {"id": document_uuid} to select from. Default []
Returns: A list of retrieved objects. Performs a semantic search and fulltext search and combine the two results. The scores are combined using Reciprocal Rank Fusion(RRF). Returns the Score (semantic similarity score), Full-text score (BM25 score), and RRF score (the final score used for getting the top N). 
warehouse = client.get_warehouse("warehouse-id")

response = warehouse.retrieve_with_hybrid_search(
    query="financial reports from Q4",
    n_objects=10,
    indexes=["chunks"],  # Optional, defaults to ["chunks"]
    reformulate_query=True, # Optional, defaults to False
    enrich_with_chunks=True, # Optional, defaults to False
    included_tags = [{"parent": "job-category", "children": "data-engineer"}],
    excluded_tags = [{"parent": "job-category", "children": "data-engineer"}],
    rerank=True, # Optional, defaults to False
)

for result in response:
    print(f"Score: {result.get('score', 'N/A')}")
    print(f"Content: {result.get('content', '')[:100]}...")
Parameters:
  • query (str): Natural language search query
  • n_objects (int, optional): Number of results. Default: 10
  • indexes (list[str], optional): Specific indexes. Default: [“chunks”]
  • reformulate_query (bool, optional): Whether to perform HyDE query reformulation before retrieving objects. Default: False
  • enrich_with_chunks (bool, optional): Whether to enrich the retrieved objects with their original chunks. Default: False
  • included_tags (list[dict], optional): Tags to include in the search. Default: []
  • excluded_tags (list[dict], optional): Tags to exclude in the search. Default: []
  • rerank (bool, optional): Whether to perform reranking on retrieved objects. Default: False
  • included_objects (list[dict], optional): Objects to include in the retrieval. A list of [{"name": entity_name}]. Default: []
  • excluded_objects (list[dict], optional): Objects to exclude in the retrieval. A list of [{"name": entity_name}]. Default: []
  • selected_documents (list[dict], optional): List of {"id": document_uuid} to select from. Default []

​
Cypher Queries

​
Generate a cypher query from a user instruction

instruction = "All the documents which contains chunks about financial data"

response = warehouse.generate_cypher_query(instruction)
cypher_query = response["generated_cypher_query"]

​
Execute Neo4j Cypher queries against the warehouse graph database:

cypher_query = """
MATCH (d:Document)-[:CONTAINS]->(c:Chunk)
WHERE d.title CONTAINS 'financial'
RETURN d.title, c.content, d.upload_date
ORDER BY d.upload_date DESC
LIMIT 10
"""

response = warehouse.retrieve_with_cypher(cypher_query)
results = response["retrieval_results"]

for result in results:
    print(f"Document: {result['d.title']}")
Parameters:
  • cypher_query (str): Valid Cypher query string
Returns: Dict with results under “retrieval_results” key

​
Execute a templated Cypher query against the warehouse graph database:

query_template = """
MATCH (n:Object)-[r]-(m:Chunk)
WHERE n. location = $job_location
LIMIT 100
WITH collect(DISTINCT n) + collect(DISTINCT m) AS nodes,
    collect(DISTINCT r) AS relationships
RETURN nodes, relationships
LIMIT 10
"""
template_variables={"job_location": "San Francisco"}

results = warehouse.retrieve_with_templated_query(query_template=cypher_query, template_variables=template_variables, enrich_with_chunks=True)

for result in results:
    print(f"Result: {result}")
Parameters:
  • query_template (str): Valid Cypher query string. This query string can include variables ($something)
  • template_variables: (dict) A dictionary of {variable_name: variable_value}
  • enrich_with_chunks: (bool) If true, objects are enriched with the content of the chunk they derive from
Returns: A list of dicts, each representing objects or chunks.

​
Generating answers

Generate answers to questions using different search strategies to retrieve relevant context. Generate an answer using semantic search to retrieve relevant context: 
warehouse = client.get_warehouse("warehouse-id")

answer = warehouse.generate_answer_with_semantic_search(
    question="What were our Q4 financial results?",
    query=None,  # Optional, defaults to question if not provided
    n_objects=10,  # Optional, defaults to 10
    indexes=["chunks"],  # Optional, defaults to ["chunks"]
    reformulate_query=True, # Optional, defaults to False
    enrich_with_chunks=False,  # Optional, defaults to False
    included_tags=[{"parent": "department", "children": "finance"}],  # Optional
    excluded_tags=[],  # Optional
    rerank=True, # Optional
)

print(f"Answer: {answer}")
Parameters:
  • question (str): The question to answer
  • query (str, optional): Custom search query for context retrieval. Defaults to question if not provided
  • n_objects (int, optional): Number of context objects to retrieve. Default: 10
  • indexes (list[str], optional): Specific indexes to search. Default: [“chunks”]
  • reformulate_query (bool, optional): Whether to perform HyDE query reformulation before retrieving objects. Default: False
  • enrich_with_chunks (bool, optional): Whether to enrich retrieved objects with original chunks. Default: False
  • included_tags (list[dict], optional): Tags to include in the search. Default: []
  • excluded_tags (list[dict], optional): Tags to exclude from the search. Default: []
  • rerank (bool, optional): Whether to perform reranking on retrieved objects. Default: False
Returns: Generated answer string or None if generation fails

​
Using Full-text Search

Generate an answer using full-text search to retrieve relevant context: 
warehouse = client.get_warehouse("warehouse-id")

answer = warehouse.generate_answer_with_full_text_search(
    question="What were our Q4 financial results?",
    query=None,  # Optional, defaults to question if not provided
    indexes=["chunks"],  # Optional, defaults to ["chunks"]
    reformulate_query=True, # Optional, defaults to False
    n_objects=10,  # Optional, defaults to 10
    enrich_with_chunks=False,  # Optional, defaults to False
    included_tags=[{"parent": "department", "children": "finance"}],  # Optional
    excluded_tags=[],  # Optional
    rerank=True, # Optional
)

print(f"Answer: {answer}")
Parameters:
  • question (str): The question to answer
  • query (str, optional): Custom search query for context retrieval. Defaults to question if not provided
  • indexes (list[str], optional): Specific indexes to search. Default: [“chunks”]
  • reformulate_query (bool, optional): Whether to perform HyDE query reformulation before retrieving objects. Default: False
  • n_objects (int, optional): Number of context objects to retrieve. Default: 10
  • enrich_with_chunks (bool, optional): Whether to enrich retrieved objects with original chunks. Default: False
  • included_tags (list[dict], optional): Tags to include in the search. Default: []
  • excluded_tags (list[dict], optional): Tags to exclude from the search. Default: []
  • rerank (bool, optional): Whether to perform reranking on retrieved objects. Default: False
Returns: Generated answer string or None if generation fails Generate an answer using hybrid search (combining semantic and full-text search) to retrieve relevant context: 
warehouse = client.get_warehouse("warehouse-id")

answer = warehouse.generate_answer_with_hybrid_search(
    question="What were our Q4 financial results?",
    query=None,  # Optional, defaults to question if not provided
    indexes=["chunks"],  # Optional, defaults to ["chunks"]
    reformulate_query=True, # Optional, defaults to False
    n_objects=10,  # Optional, defaults to 10
    enrich_with_chunks=False,  # Optional, defaults to False
    rrf_k=60,  # Optional, RRF constant for score combination
    included_tags=[{"parent": "department", "children": "finance"}],  # Optional
    excluded_tags=[],  # Optional
    rerank=True, # Optional
)

print(f"Answer: {answer}")
Parameters:
  • question (str): The question to answer
  • query (str, optional): Custom search query for context retrieval. Defaults to question if not provided
  • indexes (list[str], optional): Specific indexes to search. Default: [“chunks”]
  • reformulate_query (bool, optional): Whether to perform HyDE query reformulation before retrieving objects. Default: False
  • n_objects (int, optional): Number of context objects to retrieve. Default: 10
  • enrich_with_chunks (bool, optional): Whether to enrich retrieved objects with original chunks. Default: False
  • rrf_k (int, optional): Constant used in Reciprocal Rank Fusion algorithm. Default: 60
  • included_tags (list[dict], optional): Tags to include in the search. Default: []
  • excluded_tags (list[dict], optional): Tags to exclude from the search. Default: []
  • rerank (bool, optional): Whether to perform reranking on retrieved objects. Default: False
Returns: Generated answer string or None if generation fails