Reference
You never call these directly — you talk in plain language and Talk2Graph chooses the right action. This page documents what those actions are, the query language underneath, and the settings that govern them, so you understand what's happening and can phrase requests precisely.
Capabilities
Talk2Graph exposes the following capabilities to the assistant. Every one operates on the graph session you have open.
Describe the session
Returns the graph's schema and metadata — node and edge counts, column names, active encodings and filters, title. Optionally includes a few sample rows.
Triggered by: What's in this graph?, What columns are available?
Query the graph
Runs a read-only Cypher query to filter, traverse, count, or list nodes and edges. Can return nodes, edges, both, or just counts. Tabular results appear inline in the chat as a scrollable table (up to 100 rows).
Triggered by: Show me all Author nodes, How many nodes have degree > 10?, Top 10 by pagerank
Create a collection (color a subset)
Colors the nodes (or edges) matching a filter. Takes a color as either an exact hex code or a named palette, plus a name and one-line description. See Collections for the full picture and the palette list.
Triggered by: Color the fraud accounts red, Highlight the most influential nodes
Manage collections
List all collections (with their names, colors, and stacking order); reorder them (which renders on top); rename or recolor one; or delete one. The built-in global background layer can't be deleted.
Triggered by: What collections are on the graph?, Bring the red set to the front, Rename that to Key Accounts and make it blue, Delete the pagerank collection
Reset collections
Clears all collections at once, returning to the default view.
Triggered by: Reset the colors, Clear everything
Summarize communities
Runs community detection (Louvain) and returns a read-only text report of the largest clusters — each
cluster's size and its distinguishing traits (attribute values that are over-represented,
shown as value: X% in cluster vs Y% overall), plus a headline. It does not color the graph; to
visualize a cluster, follow up and ask to color it. You can optionally name which columns to break the
communities down by.
Triggered by: Summarize the top communities, What distinguishes the clusters?, Break the communities down by region
List sessions
Lists your own open visualization sessions.
Triggered by: What sessions do I have open?
Cypher: the query language underneath
When you ask for something, Louie writes a read-only Cypher query. Graphistry runs a subset of openCypher by translating it into GFQL, Graphistry's graph query language. You never have to write it, and when the assistant references the query, it's highlighted as inline code so you can see what ran and refine it.
Supported: MATCH / WHERE / RETURN / WITH / ORDER BY / LIMIT / DISTINCT, aggregates such as
count and collect, and graph algorithms through CALL graphistry.degree() and
CALL graphistry.igraph.<algorithm>() (PageRank, betweenness, community detection, and others).
Each query uses at most one relationship, in one direction. Write operations
(CREATE / MERGE / DELETE / SET) are never run.
For example, "color the top influencers" becomes a query that ranks nodes by PageRank and collects the top slice:
CALL graphistry.igraph.pagerank() YIELD nodeId, pagerank
WITH nodeId, pagerank ORDER BY pagerank DESC LIMIT 100
RETURN collect(nodeId) AS ids… and the assistant colors those nodes as a collection.
Limits
The current boundaries, so you know what to expect.
| Limit | Value |
| Collections per session | 10 (oldest dropped past the cap) |
| Query row limit | 50 rows (20 unprojected) |
| Inline result table | 100 rows |
| Uploaded skill size | 20,000 characters |
| Tool timeout | 60 seconds |