Example prompts

Talk2Graph understands plain language — you do not need special syntax. The examples below are grouped by what you're trying to do. They are patterns, not magic words: rephrase them freely to match your own graph and columns. Where a prompt refers to a column (like threat_score or account_type), swap in a column that exists in your data.

 

Understand the graph

Ask about the shape and schema of what you're looking at.

  • What's in this session?
  • Describe this graph
  • What columns do the nodes have?
  • How many nodes and edges are there?
  • Show me a few sample rows
 

Find, filter, and count

Show, list, or count nodes and edges matching a description — including multi-hop traversals. Results that come back as tables are shown inline in the chat as a collapsible table you can scroll.

  • Show me all Author nodes
  • List nodes where status is active
  • How many nodes have degree greater than 10?
  • Top 10 accounts by pagerank
  • Which workflow_family spans the most business_units?
  • Show the full lineage of dashboard_00490 (multi-hop traversal)
  • Find everything connected to account C17693 within two hops
 

Color and highlight

Paint a subset a color to make it stand out. Each coloring becomes a collection you can manage afterward. You can give an exact color (hex) or a named palette.

  • Color all Author nodes red
  • Highlight fraud accounts in #00FFFF
  • Highlight the PII-flagged assets using the Blues palette
  • Color nodes where amount is between 1000 and 5000 orange
  • Color the edges between C17693 and its neighbors green

Centrality — no algorithm name needed

You don't have to know graph theory. Describe the idea and Talk2Graph picks the right measure and colors the top nodes for it:

If you say……it uses
central, influential, important, key, hubPageRank (colors the top ~20% most central)
bridge, broker, connector, chokepointBetweenness
prestige, authority, most referenced, citedAuthority score
  • Highlight the most influential accounts
  • Color the bridge nodes purple
  • Show the most authoritative pages
 

Detect communities

Run community detection and get a plain summary of each cluster — its size and, importantly, what makes it distinct (the attributes that are over-represented in it). This produces a text summary; it does not color the graph on its own, but you can follow up and ask to color a specific community.

  • Summarize the top communities
  • What clusters exist in this graph?
  • Show me the 3 biggest communities and what distinguishes them
  • Break the clusters down by sentiment
  • Now color community 1 (follow-up after a summary)
 

Manage what's on screen

Adjust or clean up the colorings you've created.

  • What collections are on the graph right now?
  • Bring the bridge-accounts collection to the front
  • Rename that collection to "Key Influencers" and make it blue
  • Remove the pagerank collection
  • Reset the colors / Clear all collections
 

Refining a result

Conversation is stateful — you can build on the previous answer. If a coloring or filter isn't quite right, just describe the change:

  • Same thing, but only where amount > 1000
  • Make that a darker blue
  • Now also include the edges