Knowledge graph on the lakehouse
Define entities and relationships over your existing lakehouse tables. Query them with PuppyGraph without copying your data into a separate graph database.
Connected context for your applications
Query relationships · retrieve context
PuppyGraphKnowledge graph on your lakehouse
Graph query engine · Gremlin · openCypher
Define entities and relationships
Iceberg
Delta Lake
HudiData stays in your lakehouse. No separate graph data copy.
Built for your lakehouse ecosystem
Bring graph queries to your lakehouse tables through the catalogs and storage behind them.
Apache IcebergDelta LakeApache Hudi
Amazon S3 Tables
Microsoft OneLake
Google Cloud Lakehouse
Databricks
Snowflake
Cloudera Open Data Lakehouse
Apache Polaris
Nessie
Unity Catalog
Hive Metastore
Amazon S3
MinIO
Google Cloud Storage
Azure Blob StorageThe relationship gap
Your lakehouse brings enterprise data together. Define business entities, their properties, and the meaning of their relationships in an explicit graph model.
A foundation already in place. Your lakehouse stores and manages data from across the business, providing a foundation for analytics and applications.
A connected world within the data. Tables, fields, and identifiers represent business entities and their relationships, even when they are not modeled as a graph.
A graph model makes the difference. Having the data does not by itself provide a queryable knowledge graph. An explicit model and graph query engine let you work with it as entities, relationships, and paths.
Beyond rows and columns
A knowledge graph gives these entities and relationships an explicit model, so you can understand and query them together.
Separate compute from storage
A traditional graph database brings data into its own storage and compute system. PuppyGraph leaves the underlying data in the lakehouse and supplies the model and compute needed to query it as a graph.
IcebergDelta LakeHudiIcebergDelta LakeHudiPuppyGraphgraph model + query engineFrom tables to graph
Define the business model in a graph schema. Let PuppyGraph’s graph query engine plan and execute queries over the connected data. The model describes the knowledge; the engine makes it queryable.
Configure a catalog connection with the endpoint and access settings for your lakehouse. This makes its tables available for graph modeling.
Apache IcebergSelect tables from the connected catalog. Map them to typed nodes, relationships, and properties in a graph schema without copying the source data.
PuppyGraph uses its query engine and compute resources to run openCypher or Gremlin patterns, multi-hop paths, aggregations, and algorithms.
Which products depend on Supplier S42?
openCypher
MATCH (p:Product)-[:USES]-> (:Component)-[:SUPPLIED_BY]-> (s:Supplier {id: "S42"})RETURN DISTINCT p.id
Follow each product’s components to their supplier. Return the IDs of products that depend on S42, without duplicates.
Flexible model
Choose the entities, relationships, and properties that fit your business questions. Edit the schema as those needs change, and export it as JSON for version control. Different model definitions can draw on the same underlying tables.
Explore graph modeling →Dedicated query engine
PuppyGraph supports distributed query execution in cluster mode, parallelizing queries across compute nodes. Size graph compute for the latency and throughput your applications need, and validate performance against your workload.
Explore cluster deployment →AI agents
Give AI agents graph query tools to follow explicitly modeled relationships and retrieve connected context from your lakehouse data. Use that context to ground answers in GraphRAG workflows.
Explore GraphRAG and AI agents →What connected knowledge unlocks
From security investigations and AI context to plant operations and incident response, teams use PuppyGraph to make the relationships in their lakehouse data queryable.
Follow the investigation, not a predefined dashboard path.
Prevalent AI added graph traversal to the cyber knowledge graph its data fabric already maintained on Iceberg. Analysts can explore relationships across people, hosts, and vulnerabilities through self-service dashboards and Navigator, its conversational analytics interface.
Apache IcebergPuppyGraph queries the existing entity-resolved tables, with no separate graph store to load and synchronize.
“We found that GPEs like PuppyGraph—especially in cached mode—excel at analytical and summarization use cases, outperforming most graph databases by a wide margin.”
Namith CholappurathHead of Architecture, Prevalent AI
Give AI connected enterprise context.
AMD added a knowledge graph layer over Iceberg to connect context about users, devices, applications, and other enterprise entities. Multi-hop graph queries supply connected enterprise context to chatbots and agents.
Apache IcebergGraph access alongside the existing data stack, without a separately loaded and synchronized graph store.
“This approach enables enterprises to bridge structured data from various sources and build relationships enabling AI for smarter decision-making.”
Rajdeep SenguptaSenior Director Application & Systems Engineering, AMD
Connect equipment failures to their operational context.
A Baker Hughes study used PuppyGraph to connect equipment, incidents, root-cause analyses, bad-actor records, and work orders into a knowledge graph over Databricks Delta tables. AI agents generated graph queries from plain-language questions to identify high-loss equipment and recurring reliability problems across four plants.
Databricks · Delta LakeFive datasets queried as a graph without moving them into a separate graph database.
“PuppyGraph functions as a zero-ETL graph analytics engine that allows direct querying of data from sources like Databricks without requiring data movement.”
Ryan ElsingaSenior Data & AI Engineer, Baker Hughes
Investigate service failures and understand their impact.
Coinbase combined tracing data with service ownership information to build a service dependency graph. PuppyGraph queries helped its RCA Bot expand investigations to related services, while engineers could follow dependencies directly to understand which services and products might be affected.
Databricks · Delta LakePuppyGraph queried service relationships over data in Databricks, keeping graph compute separate from storage so the data remained available for other analyses.
“PuppyGraph is a very interesting graph query engine. It doesn't require us to load or ETL any of the data into a specialized or proprietary database storage layer for graphs. PuppyGraph queries all our transactional and crypto data directly and performs all kinds of graph-based exploration and aggregation. It's so powerful, and our users really enjoy this type of flexibility.”
Eric SunSr. Manager of Data Platform, CoinbaseKnowledge graph modeling
RDF & SPARQL are not the only way to build a knowledge graph. PuppyGraph uses the labeled property graph model, with openCypher and Gremlin to query the graph. Model the concepts and relationships in your domain ontology to bring explicit business semantics to your lakehouse data.
Entities become nodes. Named, directed edges connect them. Both carry properties.
Product, Component, and Supplier express domain concepts. USES and SUPPLIED_BY give relationships meaning. Properties describe the entities and how they connect.
Two graph query languages for finding patterns, following relationships across multiple hops, and filtering or aggregating results.
Example: find products that depend on Supplier S42.
A declarative language: describe the graph pattern you want to find.
MATCH (p:Product)-[:USES]->
(:Component)-[:SUPPLIED_BY]->
(:Supplier {id: "S42"})
RETURN DISTINCT p.idA graph traversal language: compose steps to navigate and process connected data.
g.V().has("Supplier", "id", "S42")
.in("SUPPLIED_BY").in("USES")
.hasLabel("Product")
.values("id").dedup()Start with your own data
Download PuppyGraph to start building, book a demo to explore your use case, or join Slack to get support directly from our engineering team.
$ docker run -p 8081:8081 \
-p 8182:8182 -p 7687:7687 \
puppygraph/puppygraph:latestPulling puppygraph/puppygraph:latestStarting PuppyGraphWeb console listening on 0.0.0.0:8081Gremlin endpoint listening on 0.0.0.0:8182PuppyGraph is running# no pipelines · no second database · no data movement