Ask your graph: Introducing Linkurious Copilot  

Finding answers in complex networks of connected data isn't always straightforward, especially when analyzing them requires technical skills that not every analyst has. What often starts with a single entity can quickly turn into hours of following leads, testing hypotheses, running graph queries, applying different analytical techniques, switching between tools to gather additional context, and piecing together findings until the bigger picture begins to emerge. 

Even experienced analysts spend considerable time on these tasks, while those without advanced technical skills may struggle to explore the data independently. 

Now what if investigating complex connected data was as simple as asking questions? Imagine having an AI assistant that can explore your entire graph, perform multiple-step analysis to answer your questions, and help make sense of the findings in seconds, all that through a natural language conversation.  

Meet Copilot, the new AI-powered investigation assistant built into Linkurious Enterprise.   

Going beyond natural-language query generation, Copilot combines large language models (LLMs) with graph search, queries, and visualization to help investigators explore complex networks, ask follow-up questions, and make sense of connected data through a conversation.  

Whether you're investigating suspicious transactions, examining fraud networks, or tracing connections between entities, Copilot helps you explore data faster, uncover evidence-backed findings, and focus on what matters most. 

And because investigations require more than plausible AI-generated explanations, Copilot keeps the underlying evidence accessible and investigators firmly in control. 

Imagine you're investigating a network of real estate agents and mortgage brokers potentially involved in collusion. 

You could start by asking: 

“Which pairs of broker and realtor share the most loans?” 

Copilot examines the graph and identifies the pairs with the highest number of shared loans. You might then ask: 

“Which of these loans look suspicious?” 

Copilot analyzes the available loan details and relationships, highlighting potential red flags and explaining the reasons behind its findings. 

You could then investigate specific entities and ask “Can you provide the KYC details on this specific borrower and analyze the source of the funds used to repay the loan? 

Rather than manually building and running separate queries at every step, you can progressively refine your investigation through a natural-language conversation. 

Copilot brings several AI capabilities directly into the Linkurious Enterprise investigation experience: 

• Orient yourself faster by understanding the structure of an unfamiliar network. 

• Investigate systematically with AI-assisted analysis tailored to your objective. 

• Search entities and run graph queries using natural language. 

• Connect responses back to the graph through references to relevant entities and relationships. 

• Inspect the Cypher query behind a response for greater visibility into how Copilot queried the graph. 

• Summarise findings and identify areas for further investigation. 

• Review the evidence supporting each observation before drawing conclusions. 

For analysts, investigators and domain experts who don't work with graph query languages every day, Copilot makes sophisticated graph analysis more accessible. If you’re an experienced analyst or a data scientist, it reduces the effort involved in writing complex queries and exploring networks. In both cases, the goal is the same: spend less time retrieving information and more time analyzing evidence and advancing investigations. 

Unlike a general-purpose chatbot, Linkurious' AI assistant doesn't simply generate answers based on an LLM's existing knowledge. It combines AI with direct access to your graph, using a retrieval-augmented approach to ground responses in actual entities, relationships, and properties. 

When you ask a question, Copilot follows a multi-step process: 

1. Understand the context. Copilot considers your question alongside the graph schema, the current visualization, and any business-specific guidance configured by your administrator. 

1. Search and query the graph. It uses search tools and generates Cypher queries to retrieve relevant entities, relationships, and properties. Queries are checked against read-only restrictions before execution. 

1. Interpret and explain the results. It uses the retrieved graph data to produce a natural-language explanation, with references to graph entities where applicable. 

Depending on the question, Copilot can combine several searches and queries before responding. It also takes your conversation history into account, so you can progressively refine your analysis rather than starting over with every question. 

Administrators can also configure business context, such as internal terminology, fraud typologies, or investigation guidelines. This helps Copilot interpret questions within your organization's domain. 

Note: Copilot is currently available as a Beta feature. As with any AI assistant, responses may occasionally be inaccurate or incomplete. We recommend validating important findings against the underlying graph data before making investigative decisions. 

Using AI with sensitive data raises legitimate questions. Which model processes your information? What data is shared? Can AI access restricted information or modify your graph? 

Copilot addresses these concerns through three design principles. 

Copilot follows a Bring Your Own Model (BYOM) approach. Rather than requiring a model hosted by Linkurious, your organization connects to its own approved LLM. 

Supported options include Azure OpenAI, AWS Bedrock, direct APIs from providers such as OpenAI, Anthropic, Google, and Mistral, and self-hosted models. 

This gives organizations control over model selection, deployment arrangements, and applicable data-processing terms, including the option to keep AI processing within their own infrastructure. 

Copilot inherits the role-based access controls (RBAC) of Linkurious Enterprise. It can only retrieve graph information that the signed-in user is authorized to access. 

To answer questions, Copilot shares relevant context and retrieved data with the configured LLM, including property values when needed. It doesn't export your entire graph. Information is retrieved as needed, within the user's access permissions. 

Copilot does not use your prompts or graph data to train third-party AI models. Organizations should nevertheless review the data-processing arrangements of their chosen LLM provider. 

Copilot is read-only in its current beta version. It can search and query graph data but cannot create, modify, or delete entities and relationships. 

Analysts can inspect the generated queries and retrieved information to understand how an answer was produced. 

Already a Linkurious customer or partner? Upgrade to Linkurious Enterprise 4.3 and get Copilot up and running in minutes with a simple administrator configuration. You can also customize it with your organization's business context to help Copilot understand your terminology, data, and investigation needs. 

Contact your Customer Success Manager to learn how to enable Copilot and evaluate it in your environment. For more information, read our documentation. 

As Copilot is still in beta, we'd love to hear your feedback! Your experience and suggestions will help us shape its future and build capabilities that address real investigative challenges. 

New to Linkurious? See how graph visualization, analytics, and AI-assisted investigations can help your team make sense of connected data. Request a 30-day free trial.