Open-Sourcing the WhyHow Knowledge Graph Studio, powered by NoSQL
We are excited to announce the open-source release of the WhyHow.AI Knowledge Graph Studio (github) under the MIT license! With this release, we invite the community to explore, experiment, and customize the WhyHow platform to meet their unique needs for Agentic RAG and modular knowledge graph workflows and applications. The open source repository includes the backend logic for running the knowledge graph studio including APIs, database and retrieval logic, prompts, etc.
Unique aspects of WhyHow’s Knowledge Studio Technical Architecture
Built on top of MongoDB, the Knowledge Graph Studio is designed to accommodate the unique requirements that LLM systems, multi-Agentic RAG workflows, and structured knowledge representations require, combining all the benefits of relational data, vector storage and search, and flexible schemas. We are aiming to be database agnostic and also working with a number of other partners to bring similar capabilities to other relational and graph databases.
The key missing piece in many existing workflows is in overcoming the limitations of vector search by combining structured and unstructured search together, so that you can find semantically similar sections of information, and bring in all the linked information into the context window in a structured way.
By being able to combine Text Chunks with Triples and Nodes, we are able to perform far more accurate, complete and deterministic information workflows as seen in these various case studies here (healthcare), here (legal), or here (finance).
WhyHow’s query engine also takes in natural language questions that return the nodes, triples and linked vector chunks that are related to the answer. Part of the secret sauce of WhyHow’s query engine is that it embeds triples, and retrieves it through semantic similarity, taking into account all the relevant properties, relationships and text chunks that are linked accordingly. Embedded triples contain far richer data than simply attempting to retrieve embedded nodes.
Since WhyHow’s query engine does not utilize Text2Cypher, the results are up to 2x more accurate, as can be seen in this benchmark here, where the same Graph created with Langchain’s LLMGraphTransformer gets better retrieval responses from WhyHow’s query engine than Text2Cypher.
Why Open Source?
For us, this decision is about fostering community, collaboration, and trust. We’ve seen how transformative graph-based RAG can be for building reliable, explainable, and accurate AI systems, and by open sourcing our platform, we want to enable teams to adapt and extend it for their specific needs. We aim to empower developers, researchers, and AI enthusiasts — from startups to non-profits to enterprises — to shape every aspect of graph construction, entity resolution, querying, and more to best suit their data and use cases.
Transparency, trust, and privacy are also at the core of this release. With an open source platform, teams can now take full control to install, extend, and integrate the platform directly in their own environments if they so choose. This flexibility allows users to embed custom security measures, monitoring tools, AI systems, and databases — enabling quick, compliant deployment of knowledge graphs tailored to their needs.
from whyhow import WhyHow, Triple, Node, Relation
from random import choice
# Initialize client
client = WhyHow()
# Create workspace
workspace = client.workspaces.create(name="Students")
# Build triples
person_names = [
"Chris", "Alex", "Jordan", "Taylor", "Morgan", "Casey", "Jamie", "Sam"]
subjects = [
"Computer Science", "Math", "Physics", "Chemistry", "Biology", "History"]
triples = [
Triple(
head=Node(
name=choice(person_names),
label='Student'
),
relation=Relation(
name='studies',
properties={'semester': choice('fall','spring')}
),
tail=Node(
name=choice(subjects),
label='Subject'
)
)
for _ in range(16)
]
# Create graph
client.graphs.create_graph_from_triples(
name="Classes graph",
triples=triples,
workspace_id=workspace.workspace_id
)Key Features in the Knowledge Studio
We designed the Knowledge Graph Studio to be the simplest and most intuitive way to build RAG-native knowledge graphs, agentic memory, and other structured knowledge representations. Here are some of the features that our users love:
- JSON-Native & API-First Approach with SDK: WhyHow is built with an API-first philosophy, designed to provide maximum flexibility and integration capabilities. Our JSON-native data ingestion workflows & Python SDK makes it easy to interact with every feature programmatically, allowing developers to seamlessly integrate WhyHow into their existing systems and RAG pipelines.
- Modular, Small Graph Creation: We believe in the power of modular graphs, allowing users to create focused, self-contained knowledge representations for specific datasets or use cases. This “small graph” approach supports targeted experimentation, fine-tuning, and efficient debugging, ideal for RAG applications where precision and control are crucial.
- Wide Range of Graph Structures: Data and Use-Cases are idiosyncratic. In our Medium blog, we publish a wide range of data and graph structures (including code) that lets you see how you can use different data structures, and data transformation processes in conjunction with the WhyHow platform. The WhyHow team works closely with clients to build use-case specific data transformation pipelines that are suitable for their specific needs.
- Vector Chunks as First-Class Citizens: We treat vector chunks as core primitives in graphs, making depth, context, and explainability native to data retrieval. These chunks are visible and easily manageable within each graph, supporting high-quality, context-aware information retrieval that enriches RAG workflows.
- Multiplayer Graph Creation: the WhyHow UI enables collaborative graph building by allowing non-technical users to view, interact with, and contribute to knowledge graphs. With features like graph sharing in an intuitive interface, teams can work together to create and refine graphs.
- Human-in-the-Loop Entity Resolution & Rules: WhyHow empowers users to perform personalized, use-case-specific entity resolution through a rule-based system. Easily merge similar entities and save as reusable rules, improving graph creation over time.
For those interested in a hosted experience, our licensable enterprise platform includes an innovative UI that supports multiplayer graph creation and human-in-the-loop workflows, enabling collaborative, customizable graph construction. This interface brings domain experts and developers together to create comprehensive, reliable knowledge graphs with ease. Please reach out to us at team@whyhow.ai or schedule some time with us to learn more, get signed up, or explore installation in your own environment.
Fostering an Open, Evolving Ecosystem
As knowledge graphs become increasingly central to AI systems, our aim is to advance the core technologies that make them powerful for RAG workflows. Our platform brings an opinionated approach to triple extraction, entity resolution, graph management, schema construction, and more, building on the inherent strengths of knowledge graphs that make them so valuable for reliable and explainable AI. With our open-source release, we’re excited to share these advancements with the community, supporting the broader adoption of robust, graph-enabled AI solutions and pushing the boundaries of what graph-based RAG workflows can achieve.
As part of our open source ecosystem, check out Knowledge Table, our Multi-Document Extraction & Graph Creation tool, that allows you to easily adapt your data extraction processes and simplify graph creation.
An Invitation to Get Started
The Github repo for the Knowledge Graph Studio can be found here.
Like most OS platforms, the UI will still be closed source. If you’re interested in using the UI or to get access to our hosted beta version, please book some time with us. You can also learn more about the UI and our other features on our blog.
Whether you’re new to knowledge graphs or experienced in building Graph RAG workflows, we invite you to explore and extend the WhyHow platform. Ping us at team@whyhow.ai or schedule some time with us with any questions, to learn more about the UI, or to get signed up for the hosted platform beta.
