Ask Your Graph
š¬Ā Ask Your Graph is an AI agent that can interact with your graph to help you find insights. It can search the graph, make calculations, build models, create views, and generate plots to give you the answer you require.
šĀ Access
From the homepage: navigate to the chat icon for your specific graph from the very right āOptionsā column.
From the graph view: navigate to the āAskā toggle button located at the top right corner of your screen.
š¤Ø How does it work?
Conodeās āAsk Your Graphā uses an LLM in combination with a retrieval technique called GraphRAG to provide accurate and reliable insights in response to your queries. GraphRAG understands data relationships between data sources fused in Conode. The retrieval process can therefore search across multiple sources of data, providing richer context in its answers.
Conode vs ChatGPT
LLMs hallucinate and perform poorly on your own proprietary data when the sources are messy, siloed, and complex in structure. This isnāt an issue for Conode! Conode ingests all your data -in whatever form they are, into a connected graph, and applies a powerful combination of LLM and GraphRag to generate a well rounded answer seeped in the context of your question.
š«µš»Ā Who can use this?
We hope... everyone! ā
Conodeās Ask Your Graph interacts with your data via a representative view of your graph called a skeleton. Refer to āFor the Knowledge Graph Architectsā section below to learn more about how to design the skeleton. For optimal performance, however, we recommend that you enlist a knowledge graph architect from your company, or a member of the Conode team to quickly curate this skeleton view for you. Once you have your skeleton view, you can start chatting with the agent right away.
For the Knowledge Graph Architects
As mentioned above, the __skeleton view is used to provide an overview of the graph from which the agent will traverse through and build its answer.
To ensure optimal performance of the agent, here are some tips to bear in mind when building this the skeleton:
ā¢ Make the schema human-readable, description features concise, and keep the structure simple and followable.
ā¢ In tandem with its name, keep the skeleton lean. The view should not contain too many nodes, ideally less than 100. As a guideline, add the most useful features and a couple of example data points that connect to the features so the agent understands the structure better.
ā¢ Have isolated nodes. These are nodes not connected to any others in the skeleton view and are there to provide additional context for the LLM through the node label and content.
- Dataset Description: Describe what the dataset contains, where it comes from, what each data point/node represents, and explain features that arenāt self-explanatory.
- Graph Overview: Describe the structure of the graph, how the nodes are connected, how many features and example data nodes have been added to the skeleton.
- User Intent: What kind of questions might users ask of this data? What industry might they be from and be looking for in the data? Will users prefer chart, spatial views, or just text-based answers? Any terminology that users are likely to use that is specific to this domain?
See how weāve done so in the Enron Email Corpus and Airline Passenger Reviews here:
You can always edit the description to provide more information to the agent.
š¤Ā What can I ask?
Ask your graph anything! That might seem a little daunting, so weāve provided 2 prompts in the interface to help you get started. Just click on either and youāll get your answer.
You can Ask Your Graph in various ways:
- š§®Ā Quick Calculations: What percentage of accidents involved collisions with a curb?
- š¾ Data Retrieval: Get me 100 instances of passenger review forms collected in June 2024, and have referenced having a negative experience of the airport lounge during their experience.
- šŗļø Map Generation: Plot a spatial view of all road traffic events which took place in greater London and involved an emergency vehicle.
- š Chart Generation: Plot a chart to show the museumās monthly performance measured by average ratings from visitors.
ā FAQs
What LLM is used?
As of February 2025, Conodeās Ask Your Graph uses GPT 4.0 mini. If you wish to utilise this agent but prefer another LLM, simply reach out and let us know.
Can it speak other languages?
Oui, åÆ仄, ja, sĆ!.
Can it create charts and views?
Yes. The agent has the ability to generate charts and views, and when it does so it will send you a link to view the resulting chart in the graph. For example, you can ask for:
- Bar plots
- Geospatial views with maps
- Histograms
- Custom views with nodes relevant to your search query
We are hard at work expanding this functionality, for example introducing table views and allowing previews of the charts to be presented within the conversation window. If you have something in mind which we havenāt covered yet, simply let us know and weāll get back to you with updates as soon as we can!
Will I accidentally mess up the graph when I use the agent?
It cannot mutate the existing graph. It only has the ability to add to your graphs (e.g. new views, new features in some cases), so you need not worry about messing up the data source.
How can I find out what data is being used to answer the question?
Hover over the āData Sourceā element in the chat to read the full graph label.
To understand what data sources the agent has access to, simply ask it! It should be able to summarise itās knowledge for you. You can also ask for a full view of all your data (fair warning, this might slow down the browser!)
Can the agent consult the web to get additional information?
The agent cannot consult the web. It understands the data it was trained on, in combination with the data sources you have uploaded to Conode. To answer higher level questions, enrich the graph using the extract agent.
Can you ensure the agent's answers are accurate?
Yes, just ask for the method! The agent will explain itās thought process, and can show you the subset of data it used to generate the answer.
šĀ So.. where can I get started?
Weāre glad you asked! Check out this example Airline Passenger Reviews for an easy introduction to utilising Conodeās Ask Your Graph Agent!