RDF-StarBase

A blazingly fast RDF★ database with native provenance tracking

RDF-StarBase is a native RDF★ platform for storing, querying, and visualizing assertions about data — not just data itself. Every triple carries full provenance: who said it, when, how confident they were, and which process generated it.

Key Features

• Blazingly Fast — Built on Polars with Rust-speed DataFrame operations
• Native RDF-Star — First-class support for quoted triples and statement metadata
• Full Provenance — Every assertion tracked with source, timestamp, confidence, process
• Competing Claims — See ALL assertions, not just the "winning" one
• SPARQL-Star — Query with standard SPARQL syntax + provenance extensions
• Assertion Registry — Track data sources, APIs, and mappings as first-class entities
• REST API — FastAPI-powered web interface with interactive docs
• Graph Visualization — React + D3.js frontend for exploring knowledge graphs
• Parquet Persistence — Efficient columnar storage for analytics workloads

Why RDF-StarBase?

Traditional databases store values.
Traditional catalogs store descriptions.
RDF-StarBase stores assertions about reality.

When your CRM says customer.age = 34 and your Data Lake says customer.age = 36, most systems silently overwrite. RDF-StarBase keeps both, letting you:

• See competing claims side-by-side
• Filter by source, confidence, or recency
• Maintain full audit trails
• Let downstream systems choose which to trust

Installation

PyPI (Recommended)

pip install rdf-starbase[web]  # Include REST API dependencies

Or for minimal installation:

pip install rdf-starbase

Docker (Quickest Start)

Run the complete stack (frontend + backend + database) in a single container:

docker run -d \
  --name rdfstarbase \
  -p 8000:8000 \
  -v rdfstarbase-data:/data/repositories \
  ontusdev/rdf-starbase:latest

Open http://localhost:8000/app/ to access the web interface.

Or use docker-compose:

services:
  rdfstarbase:
    image: ontusdev/rdf-starbase:latest
    ports:
      - "8000:8000"
    volumes:
      - rdfstarbase-data:/data/repositories
    environment:
      - PYTHONUNBUFFERED=1

volumes:
  rdfstarbase-data:

Start with docker-compose up -d.

Features included in Docker:

• Monaco SPARQL editor with syntax highlighting
• Schema browser with class/property exploration
• Import/export UI for Turtle, RDF/XML, N-Triples, JSON-LD
• Interactive graph visualization with D3.js
• REST API at http://localhost:8000/docs

From Source

git clone https://github.com/ontus/rdf-starbase.git
cd rdf-starbase
pip install -e ".[dev]"

Quick Start

from rdf_starbase import TripleStore, ProvenanceContext

# Create a store
store = TripleStore()

# Add triples with provenance
prov = ProvenanceContext(
    source="CRM_System",
    confidence=0.85,
    process="api_sync"
)

store.add_triple(
    "http://example.org/customer/123",
    "http://xmlns.com/foaf/0.1/name",
    "Alice Johnson",
    prov
)

# Query with provenance filtering
results = store.get_triples(
    subject="http://example.org/customer/123",
    min_confidence=0.8
)

# Detect competing claims
claims = store.get_competing_claims(
    subject="http://example.org/customer/123",
    predicate="http://example.org/age"
)

🔍 SPARQL-Star Queries

from rdf_starbase import execute_sparql

# Standard SPARQL
results = execute_sparql(store, """
    PREFIX foaf: <http://xmlns.com/foaf/0.1/>
    SELECT ?name WHERE {
        <http://example.org/customer/123> foaf:name ?name
    }
""")

# With provenance extensions
results = execute_sparql(store, """
    SELECT ?s ?p ?o WHERE {
        ?s ?p ?o .
        FILTER_CONFIDENCE(>= 0.9)
        FILTER_SOURCE("CRM_System")
    }
""")

# ASK queries
exists = execute_sparql(store, """
    ASK WHERE {
        <http://example.org/customer/123> <http://xmlns.com/foaf/0.1/name> ?name
    }
""")  # Returns: True

Advanced Query Features

# OPTIONAL - include data when available
results = execute_sparql(store, """
    PREFIX foaf: <http://xmlns.com/foaf/0.1/>
    SELECT ?person ?name ?email WHERE {
        ?person foaf:name ?name .
        OPTIONAL { ?person foaf:mbox ?email }
    }
""")

# UNION - combine multiple patterns
results = execute_sparql(store, """
    SELECT ?entity ?label WHERE {
        { ?entity rdfs:label ?label }
        UNION
        { ?entity foaf:name ?label }
    }
""")

# BIND - computed values
results = execute_sparql(store, """
    SELECT ?product ?price ?taxed WHERE {
        ?product ex:price ?price .
        BIND(?price * 1.1 AS ?taxed)
    }
""")

# Aggregates with GROUP BY
results = execute_sparql(store, """
    SELECT ?source (COUNT(*) AS ?count) (AVG(?confidence) AS ?avg_conf) WHERE {
        ?s ?p ?o .
    }
    GROUP BY ?source
    HAVING (COUNT(*) > 10)
""")

# CONSTRUCT - generate new triples
results = execute_sparql(store, """
    CONSTRUCT {
        ?person foaf:knows ?other .
    }
    WHERE {
        ?person ex:worksAt ?company .
        ?other ex:worksAt ?company .
        FILTER(?person != ?other)
    }
""")

# INSERT DATA - add new triples
execute_sparql(store, """
    INSERT DATA {
        <http://example.org/alice> foaf:name "Alice" .
        <http://example.org/alice> foaf:age 30 .
    }
""")

# DELETE DATA - remove specific triples
execute_sparql(store, """
    DELETE DATA {
        <http://example.org/alice> foaf:age 30 .
    }
""")

# DELETE WHERE - remove matching patterns
execute_sparql(store, """
    DELETE WHERE {
        <http://example.org/alice> foaf:knows ?anyone .
    }
""")

# DELETE/INSERT WHERE - update values atomically
execute_sparql(store, """
    DELETE { ?s ex:status "active" }
    INSERT { ?s ex:status "archived" }
    WHERE { ?s ex:status "active" }
""")

# Property paths - navigate graph relationships
results = execute_sparql(store, """
    SELECT ?ancestor WHERE {
        <http://example.org/alice> foaf:knows+ ?ancestor .  # One or more hops
    }
""")

results = execute_sparql(store, """
    SELECT ?connected WHERE {
        <http://example.org/alice> (foaf:knows|foaf:worksWith)* ?connected .  # Zero or more via knows OR worksWith
    }
""")

results = execute_sparql(store, """
    SELECT ?knower WHERE {
        ?knower ^foaf:knows <http://example.org/bob> .  # Inverse: who knows Bob?
    }
""")

# Time-travel queries - query historical state
results = execute_sparql(store, """
    SELECT ?s ?name WHERE {
        ?s foaf:name ?name .
    }
    AS OF "2025-01-15T00:00:00Z"
""")

# ASK with time-travel
existed = execute_sparql(store, """
    ASK WHERE {
        <http://example.org/alice> foaf:name ?name .
    }
    AS OF "2024-06-01"
""")  # Returns: True if Alice existed on that date

📊 Named Graph Management

RDF-StarBase supports named graphs (graph containers/clusters) with full SPARQL Graph Store Protocol operations:

from rdf_starbase import execute_sparql

# CREATE GRAPH - create a new named graph
execute_sparql(store, """
    CREATE GRAPH <http://example.org/graphs/customers>
""")

# LOAD - load RDF data from a file into a graph
execute_sparql(store, """
    LOAD <file:///data/customers.ttl> 
    INTO GRAPH <http://example.org/graphs/customers>
""")

# Or load from HTTP
execute_sparql(store, """
    LOAD <https://example.org/data/products.ttl>
    INTO GRAPH <http://example.org/graphs/products>
""")

# COPY - copy all triples from one graph to another
execute_sparql(store, """
    COPY GRAPH <http://example.org/graphs/customers>
    TO GRAPH <http://example.org/graphs/customers_backup>
""")

# MOVE - move triples (copy then clear source)
execute_sparql(store, """
    MOVE GRAPH <http://example.org/graphs/staging>
    TO GRAPH <http://example.org/graphs/production>
""")

# ADD - add triples to another graph (merge)
execute_sparql(store, """
    ADD GRAPH <http://example.org/graphs/updates>
    TO GRAPH <http://example.org/graphs/main>
""")

# CLEAR - remove all triples from a graph (graph still exists)
execute_sparql(store, """
    CLEAR GRAPH <http://example.org/graphs/temp>
""")

# DROP - delete a graph and all its triples
execute_sparql(store, """
    DROP GRAPH <http://example.org/graphs/old_data>
""")

# Special graph targets
execute_sparql(store, "CLEAR DEFAULT")  # Clear default graph
execute_sparql(store, "DROP NAMED")     # Drop all named graphs
execute_sparql(store, "CLEAR ALL")      # Clear everything

# SILENT mode - don't fail if graph doesn't exist
execute_sparql(store, """
    DROP SILENT GRAPH <http://example.org/graphs/maybe_exists>
""")

# List all named graphs
graphs = store.list_graphs()
print(graphs)  # ['http://example.org/graphs/customers', 'http://example.org/graphs/products']

Querying Named Graphs

# FROM clause - restrict query to specific graph
results = execute_sparql(store, """
    SELECT ?customer ?name
    FROM <http://example.org/graphs/customers>
    WHERE {
        ?customer foaf:name ?name
    }
""")

# FROM with multiple graphs (union of datasets)
results = execute_sparql(store, """
    SELECT ?entity ?label
    FROM <http://example.org/graphs/customers>
    FROM <http://example.org/graphs/products>
    WHERE {
        ?entity rdfs:label ?label
    }
""")

# GRAPH pattern - query specific named graph in WHERE clause
results = execute_sparql(store, """
    SELECT ?customer ?name WHERE {
        GRAPH <http://example.org/graphs/customers> {
            ?customer foaf:name ?name
        }
    }
""")

# GRAPH with variable - discover which graph contains data
results = execute_sparql(store, """
    SELECT ?graph ?entity ?name WHERE {
        GRAPH ?graph {
            ?entity foaf:name ?name
        }
    }
""")

# Combined patterns - default graph + specific named graph
results = execute_sparql(store, """
    SELECT ?person ?friend ?friendName WHERE {
        ?person foaf:knows ?friend .
        GRAPH <http://example.org/graphs/profiles> {
            ?friend foaf:name ?friendName
        }
    }
""")

# FROM NAMED - specify available named graphs for GRAPH patterns
results = execute_sparql(store, """
    SELECT ?g ?s ?name
    FROM NAMED <http://example.org/graphs/customers>
    FROM NAMED <http://example.org/graphs/employees>
    WHERE {
        GRAPH ?g { ?s foaf:name ?name }
    }
""")

⭐ RDF-Star: Quoted Triples

RDF-Star allows you to make statements about statements:

# The assertion "Alice knows Bob" is claimed by Wikipedia
store.add_quoted_triple(
    subject="<<http://example.org/alice http://xmlns.com/foaf/0.1/knows http://example.org/bob>>",
    predicate="http://example.org/assertedBy",
    obj="http://dbpedia.org/resource/Wikipedia",
    provenance=prov
)

Query with SPARQL-Star:

SELECT ?who WHERE {
    << ?person foaf:knows ?other >> ex:assertedBy ?who
}

Competing Claims Detection

# Multiple systems report different ages
crm_prov = ProvenanceContext(source="CRM", confidence=0.85)
lake_prov = ProvenanceContext(source="DataLake", confidence=0.92)

store.add_triple(customer, "http://example.org/age", 34, crm_prov)
store.add_triple(customer, "http://example.org/age", 36, lake_prov)

# See all competing values
claims = store.get_competing_claims(customer, "http://example.org/age")
print(claims)
# shape: (2, 4)
# ┌────────┬──────────┬────────────┬─────────────────────┐
# │ object │ source   │ confidence │ timestamp           │
# ├────────┼──────────┼────────────┼─────────────────────┤
# │ 36     │ DataLake │ 0.92       │ 2026-01-16 03:00:00 │
# │ 34     │ CRM      │ 0.85       │ 2026-01-16 02:00:00 │
# └────────┴──────────┴────────────┴─────────────────────┘

Persistence

# Save to Parquet (columnar, fast, compressible)
store.save("knowledge_graph.parquet")

# Load back
loaded_store = TripleStore.load("knowledge_graph.parquet")

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                         RDF-StarBase                                │
├─────────────────────────────────────────────────────────────────────┤
│    React + D3.js Frontend    │     REST API (FastAPI)               │
├──────────────────────────────┼──────────────────────────────────────┤
│  SPARQL-Star Parser  │  Query Executor  │  Assertion Registry       │
├─────────────────────────────────────────────────────────────────────┤
│                    Triple Store (Polars DataFrames)                 │
├─────────────────────────────────────────────────────────────────────┤
│  Parquet I/O  │  Provenance Tracking  │  Competing Claims Detection │
└─────────────────────────────────────────────────────────────────────┘

Core Stack:

• Polars — Rust-powered DataFrames for blazing performance
• FastAPI — Modern async REST API framework
• pyparsing — SPARQL-Star parser
• Pydantic — Data model validation
• D3.js — Graph visualization
• PyArrow — Parquet persistence

Performance

RDF-StarBase leverages Polars' Rust backend for:

• Vectorized operations on millions of triples
• Lazy evaluation for query optimization
• Zero-copy reads from Parquet
• Parallel execution across cores

Web API

Start the server:

# Using uvicorn directly
uvicorn rdf_starbase.web:app --reload

# Or with the module
python -m rdf_starbase.web

Then open:

• API Docs: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc

REST Endpoints

Endpoint	Method	Description
/triples	GET	Query triples with filters
/triples	POST	Add new triple with provenance
/triples/{subject}/claims	GET	Get competing claims
/sparql	POST	Execute SPARQL-Star query
/sources	GET/POST	Manage data sources
/graph/nodes	GET	Visualization data
/graph/edges	GET	Graph edges
/stats	GET	Database statistics

🤖 AI Grounding API

A specialized API layer designed for AI/LLM consumption, separate from the UI visualization endpoints:

Endpoint	Method	Description
/ai/query	POST	Structured fact retrieval with provenance for RAG
/ai/verify	POST	Verify if a claim is supported by the knowledge base
/ai/context/{iri}	GET	Get all facts about an entity with citations
/ai/materialize	POST	Trigger reasoning and persist inferences
/ai/inferences	GET	List materialized inferences
/ai/health	GET	AI API health check

Why a Separate AI API?

Aspect	UI API (/graph/*)	AI Grounding API (/ai/*)
Consumer	D3.js visualization	LLM tool calls / agents
Response format	Nodes + edges for rendering	Facts + provenance + citations
Query pattern	Browsing, neighborhood exploration	Precise fact lookup, verification
Filtering	Limit by count, visual simplicity	Confidence threshold, freshness

Example: Grounding an AI Response

import httpx

# 1. Query relevant facts for RAG
response = httpx.post("http://localhost:8000/ai/query", json={
    "subject": "http://example.org/customer/123",
    "min_confidence": "high",  # high (>=0.9), medium (>=0.7), low (>=0.5), any
    "max_age_days": 30,        # Only recent facts
})
facts = response.json()["facts"]

# 2. Verify a claim before stating it
verify = httpx.post("http://localhost:8000/ai/verify", json={
    "subject": "http://example.org/customer/123",
    "predicate": "http://xmlns.com/foaf/0.1/age",
    "expected_object": "34",
})
result = verify.json()
if result["claim_supported"]:
    print(f"Claim verified with {result['confidence']:.0%} confidence")
elif result["has_conflicts"]:
    print("Warning: Competing claims exist!")
    print(result["recommendation"])

# 3. Get full entity context
context = httpx.get("http://localhost:8000/ai/context/http://example.org/customer/123")
entity_facts = context.json()["facts"]
related = context.json()["related_entities"]

Inference Materialization

Materialize RDFS/OWL inferences with provenance tracking:

# Run reasoning engine and persist inferred triples
response = httpx.post("http://localhost:8000/ai/materialize", json={
    "enable_rdfs": True,   # RDFS entailment rules
    "enable_owl": True,    # OWL 2 RL rules
    "max_iterations": 100,
})
print(f"Inferred {response.json()['triples_inferred']} triples")

# Query inferred facts (marked with source='reasoner')
inferences = httpx.get("http://localhost:8000/ai/inferences")
for fact in inferences.json()["inferences"]:
    print(f"Inferred: {fact['subject']} {fact['predicate']} {fact['object']}")

📋 Assertion Registry

Track data sources as first-class entities:

from rdf_starbase import AssertionRegistry, SourceType

registry = AssertionRegistry()

# Register a data source
source = registry.register_source(
    name="CRM_Production",
    source_type=SourceType.API,
    uri="https://api.crm.example.com/v2",
    owner="sales-team",
    tags=["production", "customer-data"],
)

# Track sync runs
run = registry.start_sync(source.id)
# ... perform sync ...
registry.complete_sync(run.id, records_processed=1000)

# Get sync history
history = registry.get_sync_history(source.id)

🧪 Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=src/rdf_starbase

# Format code
black src/ tests/
ruff check src/ tests/

📊 Frontend (React + D3)

cd frontend
npm install
npm run dev

Then open http://localhost:3000 (proxies API to :8000)

📚 Examples

See the examples/ directory:

• quickstart.py — Core features demonstration
• competing_claims.py — Handling conflicting data from multiple sources
• sparql_queries.py — SPARQL-Star query examples
• registry_demo.py — Assertion Registry usage

🗺️ Roadmap

✅ Completed (MVP)

• Native RDF-Star storage
• Provenance tracking (source, timestamp, confidence, process)
• Competing claims detection
• SPARQL-Star parser (SELECT, ASK, FILTER, ORDER BY, LIMIT, OFFSET)
• SPARQL-Star executor with Polars backend
• Provenance filter extensions
• Parquet persistence
• Assertion Registry (datasets, APIs, mappings)
• REST API with FastAPI
• React + D3 graph visualization

✅ Completed (Advanced Query Features)

• OPTIONAL patterns (left outer joins)
• UNION patterns (combine result sets)
• MINUS patterns (set difference)
• FILTER expressions (comparisons, boolean logic, regex, string functions)
• BIND clauses (variable assignment, expressions, functions)
• VALUES inline data
• Aggregate functions (COUNT, SUM, AVG, MIN, MAX, GROUP_CONCAT, SAMPLE)
• GROUP BY and HAVING
• CONSTRUCT queries (template-based triple generation)
• DESCRIBE queries (resource description)
• SPARQL UPDATE (INSERT DATA, DELETE DATA, DELETE WHERE, DELETE/INSERT WHERE)
• OWL reasoning (rdfs:subClassOf, owl:sameAs, owl:inverseOf, owl:TransitiveProperty)
• Property path queries (/, |, ^, *, +, ?)
• Time-travel queries (AS OF "2025-01-15T00:00:00Z")
• AI Grounding API (/ai/query, /ai/verify, /ai/context)
• Inference materialization (/ai/materialize, /ai/inferences)
• Named Graph Management (CREATE, DROP, CLEAR, LOAD, COPY, MOVE, ADD)
• FROM clause dataset specification
• GRAPH pattern queries

🔜 Next

• Trust scoring and decay

🚀 Future

• Federation across instances
• Governance workflows

📄 License

MIT License — see LICENSE for details.

🙏 Acknowledgments

• Polars — The lightning-fast DataFrame library
• RDF-Star Working Group — For the specification
• FastAPI — Modern Python web framework
• D3.js — Data visualization library
• pyparsing — Parser combinators for Python

---

RDF-StarBase — The place where enterprises store beliefs, not just data.