Skip to main content

Quickstart Guide

This guide will walk you through setting up and using Fluree's GenAI capabilities with your dataset, enabling powerful question-answering through GraphRAG data retrieval.

Prerequisites​

info

This guide assumes you have already created a Fluree account.
Please see the Account Registration guide, if you need help with this.

Introduction​

Fluree's built-in support for GenAI transforms how you interact with your structured data. While traditional AI approaches often struggle with analytical questions and can produce hallucinated or unreliable answers, Fluree's GraphRAG approach combines the best of both worlds: the intuitive interface of natural language with the precision and reliability of database queries.

What is GraphRAG and Why Does It Matter?​

GraphRAG (Graph-based Retrieval Augmented Generation) represents a significant advancement over traditional RAG systems:

  • Traditional RAG relies on vector similarity to find relevant chunks of text, which works well for document retrieval but struggles with analytical questions that require understanding relationships and performing calculations.
  • GraphRAG leverages the power of semantic graph databases to understand not just the content of your data, but its structure, relationships, and meaning. This allows it to answer complex questions involving filters, aggregations, comparisons, and multi-entity relationships with precision.

What makes Fluree's implementation special is how it eliminates the typical barriers to entry:

  • No complex setup: Other GraphRAG implementations require extensive engineering work to connect your data, LLMs, and query engines.
  • No hallucinations: Fluree's approach generates actual database queries that run against your real data, providing verifiable, accurate results every time.
  • Built-in security: Access policies are enforced at the data level, ensuring that AI interactions respect your security boundaries.

This guide focuses on one of the most immediately available and powerful patterns: native support for GraphRAG data retrieval. In just minutes, you'll transform arbitrary source data into an AI-enabled knowledge system that can answer complex analytical questions about your data concerns.

Why Semantic Graphs for GenAI?​

If you're not yet familiar with semantic graphs and their importance for RAG (Retrieval-Augmented Generation) approaches to GenAI knowledge consumption, here's why they matter:

  • Traditional vector similarity approaches face diminishing returns with analytical questions
  • Semantic graphs excel at answering questions involving filters, aggregations, entity relationships, comparators, and set grouping
  • Fluree's semantic graph approach provides provable, hallucination-free LLM query generation
note

Standing up GenAI utility around knowledge in structured data can be a significant and often expensive challenge. Fluree provides out-of-the-box enablement with minimal setup.

What You'll Accomplish​

Within minutes, this guide will walk you through:

  1. Automated migration to semantic graph data (model & instance data) from Excel & CSV files
  2. Immediate support for GenAI Q&A (answering analytical, domain-specific questions via hallucination-free LLM query generation)
  3. Seamless access policy management, allowing safe collaboration with guarantees against data leaks or AI-Agent injection threats

Let's get started!

Data Loading​

Fluree supports multiple pathways for loading data into a dataset—whether for new data production or migration from structured, semi-structured, and unstructured sources.

In this guide, we'll import data from Excel & CSV files, though Fluree supports many other data onboarding methods.

Importing Your Data​

For this guide, we'll use a Hardware Supply Chain dataset that includes information about products, materials, manufacturers, and inventory.

Data Model​

First, download the Excel model file that defines the structure of the data:

Hardware Supply Chain Model
Instance Data​

Next, download the CSV files containing the instance data:

Bill of Materials BOM Items Manufacturers Materials Material Classes ProductsWarehouse Inventory

Import Steps​

Once you've downloaded the files, follow these steps to import them into Fluree:

  1. Create a new dataset by clicking the "+" button at the top-right of the screen, then "New Dataset".
    See this guide for more details on creating a new dataset.
  2. After creating the dataset, you should automatically be taken to the Dataset page for your new dataset. Otherwise, you can navigate to your Dataset page by selecting it on your homepage.
  3. On the Dataset page, select the "Files" tab from the menu on the left side of the screen.
note

The "Files" tab allows us to import table-shaped or unstructured data to be transformed into well-modeled RDF graph data. We will be using it here to import our data model and process CSV instances.

  1. Drag the HardwareSupplyChainModel.xlsx file into the upload area (or, click the "Select Files" button and choose the file)
  2. After the file uploads, click "Process" to start the import
tip

You may also toggle on the "Automatically Process & Transact" option so that, after selecting files to upload, they will be processed without having to click the "Process" button each time.

  1. Wait for the processing to complete. If you have not toggled automatic processing on, click "Transact" when the processing is complete.
  2. Drag each of the CSV files into the upload area (or, click the "Select Files" button and choose the files)
caution

It is important to upload, process, and transact the Excel / Model file first, as the CSV instance data can only be processed correctly if the model has been established to organize the data. The data import process will fail with errors if you attempt to upload the CSV files before having introduced the model data.

  1. Process & transact each CSV file in the same way as the model file
A view of the Files UI after having uploaded & processed each file

Understanding Your Data Model​

After importing your data, Fluree automatically creates a semantic data model. To explore this model:

  1. Navigate to the "Models" tab
  2. You'll be able to use the model explorer tree to view the classes and properties of our Hardware Supply Chain data model.
  3. Clicking on a class or a property will show you its details, including any labels, comments, data types, and relationships to other classes.
  4. After having selected a class or property, you can also click "View Instances" to see the instance data that corresponds to that class or property.
A view of the hasManufacturer property in the Hardware Supply Chain data model
tip

While the data model is useful to humans for understanding the conceptual shape of our data, it is also of critical value to the LLM when we task it with generating SPARQL queries based on the question we ask it in our GenAI Chat.

A significant advantage of using Fluree's semantic graph technology for GraphRAG is that the model is machine-readable and its structure is natively intelligible to the LLM. The SPARQL query language is designed to represent queries in exact accordance with that model, so the LLM can seamlessly generate model-accurate queries without syntactical errors or hallucinations.

Exploring Your Instance Data​

If you haven't already clicked "View Instances" in the "Data Model" UI, you can also explore your instance data by navigating to the "Data" tab.

  1. Navigate to the "Data" tab
  2. The tab will default to a "Class"-oriented view, where you can select a class and either all of its known properties, or you can use the checkboxes to select specific properties to view.
  3. Selecting classes & properties in this way will populate a grid-like view of your instance data according to your selections.
  4. Clicking on any identifier in the grid will navigate you to a wiki-style "Instance Data View", where you can see all of the properties and values for that instance, as well as any relationships to other instances.
  5. Try clicking on the "Material" class. Then clicking on one of the "@id" identifiers for a specific instance. You should see links to each associated entity, as well as the properties and values for that instance.
A view of the Instance Data UI for your dataset

Initial GenAI Chat​

Now for the exciting part: interacting with your data using natural language questions!

  1. On the "Chat" navigation item, click the "+" button to start a new chat session
  2. The GenAI Chat will immediately open with a summary of your dataset's model and some helpful context suggesting the kinds of knowledge the dataset may contain.
  3. Let's ask an initial question of the dataset, such as "Find the three products in the dataset that are performing the worst with regard to competitive products on the market. Please provide some summary data about each product, including the data relevant to answering this question."
  4. In the Chat window, you'll see the AI's natural language response, which will include a list of products with the details we asked for.
  5. Clicking either the "Toggle Explanation Focus" icon in the chat response card, or the "Show Explain Panel" in the upper-right corner of the Chat UI will show you the SPARQL query generated by the LLM to answer your question, as well as the exact data response that the LLM proceeded to summarize.
info

This is a great way to prove the value of Fluree's GraphRAG-based approach to GenAI knowledge production.

Other GenAI approaches may produce natural language responses that seem accurate and correct. However, it is often impossible in those solutions to know how the LLM produced its answer from your data. If the answer was produced by vector search or only by the LLM with no data-retrieval, it is nearly impossible to answer questions involving analytical factors (e.g. in this question, we asked a question requiring a comparison of numerical values and an ordering based on that comparison. This is entirely beyond the capacities of vector-search or LLM Q/A on their own.)

With GraphRAG, we answer questions by querying the dataset itself. The "Explain" panel shows you exactly how the LLM produced its answer, and the SPARQL query can be re-used to produce the exact same answer.

A view of the Chat UI for your dataset

None of this would be possible without the critical, machine-readable information expressed by our semantic data model or without the seamless expression of model-related questions via SPARQL query syntax.

Try asking more complex analytical questions that might be typical for business analysts in the Hardware Supply Chain domain:

  • Of the materials used for assembling the 4-Bay Network-Attached Storage (NAS) product, which of these materials could be made significantly more affordable if we were to replace them in the BOM for this product with comparable materials? Please ensure you return results starting with the materials that could be the most drastically improved by replacing with an alternate material.
  • The material, INTERFACE SIEMENS CN 6ES71511AA060AB0, does seem very expensive. I want to know everything I can about any alternative material that could be used in its place. Please return the list of those materials with the lowest price material first.
  • Which 10 materials are at most risk of being out of stock? Tell me everything about those materials, including their inventory information.

Empowering GraphRAG with Access Policy​

One of Fluree's key advantages is its ability to enforce access policies at the data level. Let's introduce some basic policies to illustrate this.

  1. Go to the "Policies" link in the left navigation bar

  2. Click the "+" button to create a new "Policy Group". Let's call it "ProductAnalyst".

  3. Fill out the form to define this policy group. For now, you can leave the "Policies" section empty, as we have yet to add any relevant policies.

  4. Click the "Policies" tab to the right of the currently-highlighted "Policy Groups" tab.

  5. Click the "+" button to create a new policy.

  6. We'll start by adding a blanket "Read All" policy with the following information:

    • Policy Name: Read All
    • Description: Default read-all policy to be supplemented with more specific access denial policies.
    • Does this policy ALLOW or DENY visibility/access?: Allow
    • What action does this policy affect?: Read only
    • This policy will allow read access on...: All data
    • Policy Groups: ProductAnalyst

  7. You'll see a human summary of the policy and can proceed to click "Save Policy"

  8. Now, let's add a policy that denies access to reading data relating to inventory numbers or material unit pricing.

    • Policy Name: Deny Inventory & Pricing
    • Description: Deny access to inventory numbers and material unit pricing
    • Does this policy ALLOW or DENY visibility/access?: Deny
    • What action does this policy affect?: Read only
    • This policy will deny read access on...: Select Properties...
    • Deny read access to the properties...": type quantityUnitsAvailable and select the property from the dropdown. Do the same for materialUnitPrice
    • Policy Groups: ProductAnalyst

  9. Click "Save Policy" to save the policy

note

This is really an arbitrary example, and you should feel free to create more interesting or comprehensive policies by reading more on Fluree's Data-Centric Access Policy Management here

  1. Now, we should be able to view both our "ProductAnalyst" policy group and the policies we created & associated with that group.

Let's now return to the "Chat" UI. This time, let's ask the question...

What are all materials associated with the SAS Host Bus Adapter (HBA) product. Please return all data associated with each material, keeping in mind that entities may not have data on each associated property.

The first time we ask it, because we have not asked the question as a particular "Policy Group", our response should include data related to material unit prices and units available for that material.

Now, however, let's select "ProductAnalyst" from the "Run as policy group" dropdown near the top of the Chat UI, and then proceed to ask the same question once more.

info

Now when you or collaborators use the dataset or its GenAI chat capabilities, it is possible to ensure compliance with strict & arbitrarily-granular access controls, while still seamlessly enabling powerful analytics and knowledge consumption.

Summary & Next Steps​

Congratulations! You've just unlocked powerful GenAI capabilities with your Fluree dataset. Let's recap what you've accomplished:

  • Seamless Data Migration: You've transformed ordinary Excel and CSV files into a rich, semantic graph database that machines and AI can truly understand.
  • Instant AI-Powered Analytics: You've enabled sophisticated question-answering capabilities that go far beyond simple keyword searches, allowing for complex analytical queries using natural language.
  • Hallucination-Free Knowledge: Unlike traditional GenAI approaches, your system now provides verifiable, accurate answers with transparent query generation you can inspect.
  • Granular Access Control: You've implemented data-centric security policies that protect sensitive information while still enabling powerful analytics.

What makes this especially powerful is that you've accomplished all of this without writing a single line of code or spending weeks on complex integrations. The GraphRAG approach you've implemented represents the cutting edge of GenAI for structured data, combining the analytical power of graph databases with the intuitive interface of natural language.

Where to Go From Here​

Now that you have this foundation in place, consider these exciting next steps:

  • Create Datasets from Your Own Data: Use the same data import process and this guide to create datasets from your own data sources.
  • Create Custom Policies: Design more sophisticated access policies tailored to different roles in your organization with the help of this guide.
  • Integrate with Applications: Use Fluree's API capabilities to embed these GenAI features directly into your business applications.
  • Explore Advanced Queries: Challenge the system with increasingly complex analytical questions to discover the full extent of its capabilities.
  • Build Automated Workflows: Consider how you might use the query generation capabilities to automate regular business intelligence reporting.

The system you've built isn't just a demo—it's a production-ready foundation for transforming how your organization interacts with data. By combining semantic graph technology with state-of-the-art AI, you've created a knowledge system that's greater than the sum of its parts.