Skip to main content

Applications API

Prometheux Chain offers powerful tools for working with reasoning and knowledge graphs.

Graph RAG​

The Graph RAG API performs GraphRAG (Graph Retrieval-Augmented Generation) operations, combining graph reasoning with retrieval-augmented generation to answer questions using both structured data and language models.

Query Endpoint​

import prometheux_chain as px

result = px.graph_rag(
    project_id="my_project_id",
    question="What companies are located in California?",
    rag_concepts=[
        {"concept": "company", "field_to_embed": "name"},
        {"concept": "location", "field_to_embed": "city"}
    ],
    graph_selected_concepts=["company", "location"],
    top_k=5
)

Function Signature

def graph_rag(
    project_id, 
    question, 
    graph_selected_concepts=None, 
    graph_available_concepts=None, 
    rag_concepts=None, 
    rag_records=None, 
    project_scope="user", 
    llm=None, 
    top_k=5
)

Parameters

ParameterTypeRequiredDescription
project_idstrYesThe project identifier
questionstrYesThe question to answer
graph_selected_conceptslistNoConcept names to directly execute for graph operations
graph_available_conceptslistNoConcept names available to the LLM orchestrator for automatic selection
rag_conceptslistNoList of dicts with concept and field info for RAG operations
rag_recordslistNoList of retrieved records for RAG operations (external embeddings)
project_scopestrNoThe scope of the project. Defaults to "user"
llmdictNoLLM configuration dictionary
top_kintNoNumber of top results to retrieve. Defaults to 5

GraphRAG Modes​

GraphRAG supports different modes for both embeddings and graph concepts:

Embedding Modes​

  1. Internal Embeddings: The system performs embedding-based retrieval internally by specifying which concept and field to embed using rag_concepts (SDK) or rag.embedding_to_retrieve (REST).

  2. External Embeddings: You provide the results of embedding-based retrieval directly using rag_records (SDK) or rag.embedding_retrieved (REST).

  3. No RAG: Omit the rag section entirely to disable RAG and only use graph concepts.

Graph Concept Modes​

  1. Explicit Graph Concepts: You explicitly specify which graph concepts to run using graph_selected_concepts (SDK) or graph.selected_concepts (REST).

  2. Implicit Graph Concepts: The LLM-based orchestrator automatically decides which concepts to run based on the question and available concepts. If no concepts are specified, the orchestrator chooses from all concepts in the project.

Examples​

Internal Embeddings + Explicit Graph Concepts​

import prometheux_chain as px

# Full control over both embedding and graph operations
result = px.graph_rag(
    project_id="my_project_id",
    question="What companies are located in California?",
    rag_concepts=[
        {"concept": "company", "field_to_embed": "name"},
        {"concept": "location", "field_to_embed": "city"}
    ],
    graph_selected_concepts=["company", "location"]
)

External Embeddings + Implicit Graph Concepts​

import prometheux_chain as px

# Provide your own embeddings, let orchestrator choose graph concepts
rag_records = {
    "company": [["Apple", "Google", "Microsoft"]],
    "location": [["Redwood City, CA", "Mountain View, CA", "Redmond, WA"]]
}

result = px.graph_rag(
    project_id="my_project_id",
    question="Which companies are in California?",
    rag_records=rag_records
    # No graph parameters - orchestrator will choose from all project concepts
)

Internal Embeddings + Implicit Graph Concepts​

import prometheux_chain as px

# Control embeddings, let orchestrator choose graph concepts
result = px.graph_rag(
    project_id="my_project_id",
    question="Find companies in California",
    rag_concepts=[
        {"concept": "location", "field_to_embed": "city"}
    ]
    # No graph parameters - orchestrator will choose from all project concepts
)

Graph Only (No RAG)​

import prometheux_chain as px

# Execute graph concepts without RAG
result = px.graph_rag(
    project_id="my_project_id",
    question="Show me all companies",
    graph_selected_concepts=["company"]
    # No rag_concepts - RAG is disabled
)

Complete Workflow Example​

Use Case: Employee recommendation system that finds the most suitable employees for new projects by combining semantic search of project topics with graph analysis of team-employee-project relationships.

import prometheux_chain as px
import os

# Set up authentication and configuration
os.environ['PMTX_TOKEN'] = 'my_pmtx_token'
px.config.set('JARVISPY_URL', "https://api.prometheux.ai/jarvispy/my-org/my-user")

# Create a project
project_id = px.save_project(project_name="graphrag_demo")

# Define and save concepts for contributions
concept_logic = """
contributions_input("Phoenix","graph RAG for internal knowledge discovery; knowledge graphs",1001).
contributions_input("Phoenix","graph RAG for internal knowledge discovery; knowledge graphs",1002).
contributions_input("Atlas","hybrid retrieval; vector search; semantic search platform",1002).
contributions_input("Atlas","hybrid retrieval; vector search; semantic search platform",1005).

contributions(Project_name,Project_topic,Employee_id) :- 
    contributions_input(Project_name,Project_topic,Employee_id).

@output("contributions").
"""
px.save_concept(project_id=project_id, concept_logic=concept_logic)

# Define teams and employees
concept_logic = """
teams_employees_input("Research",1001,"Alice Nguyen").
teams_employees_input("Research",1003,"Carol Bianchi").
teams_employees_input("Engineering",1002,"Bob Rossi").
teams_employees_input("Engineering",1005,"Eve Marino").

teams_employees(Team_name,Employee_id,Employee_name) :- 
    teams_employees_input(Team_name,Employee_id,Employee_name).

@output("teams_employees").
"""
px.save_concept(project_id=project_id, concept_logic=concept_logic)

# Define projects concept
concept_logic = """
projects(Project_name,Project_topic) :- 
    contributions(Project_name,Project_topic,Employee_id).

@output("projects").
"""
px.save_concept(project_id=project_id, concept_logic=concept_logic)

# Define closest employee concept
concept_logic = """
node("project",Project_name,[]) :- contributions(Project_name,Project_topic,Employee_id).
node("employee",Employee_id,[Employee_name]) :- teams_employees(Team_name,Employee_id,Employee_name).

edge("contributed_to",Employee_id,Project_name) :- contributions(Project_name,Project_topic,Employee_id).

edge_undirected(Y,X) :- edge(Type,X,Y).
edge_undirected(X,Y) :- edge(Type,X,Y).

path(X,Y,Dist,Path) :- edge_undirected(X,Y), projects(X,Project_topic), Dist = 1, Path = [Y].

neighbor(Type,Id,Dist,Source,Path) :- path(Source,Id,Dist,Path), node(Type,Id,Attributes).

closest_employee_to_project(Id,Name,Distance) :-
    neighbor("employee",Id,Dist,Source,Path),
    node("employee",Id,Name),
    Distance = mmin(Dist).

@output("closest_employee_to_project").
@post("closest_employee_to_project","orderBy(3)").
"""
px.save_concept(project_id=project_id, concept_logic=concept_logic)

# Perform GraphRAG query
rag_result = px.graph_rag(
    project_id=project_id,
    question="I'm kicking off a project on knowledge graphs for data governance, which employees do you suggest?",
    rag_concepts=[{"concept": "projects", "field_to_embed": "Project_topic"}],
    graph_available_concepts=["closest_employee_to_project"],
    top_k=3
)

print(f"Answer: {rag_result}")