Any network, pasted as an edge list, read for real by the browser first
An org chart, a software dependency graph, a supply chain, a citation network, a social graph — paste the relationships as one pair per line. A free engine in this browser computes the exact structure: degree, connected components, cut-vertices and bridges, cycles, clustering coefficient, closeness and betweenness centrality, diameter. Then pick a lane.
The org-chart example ships with a saved model run for both lanes, so you can see the whole thing — the structural read, the handoff, and the decision brief — without signing in and without spending a credit.
Graph Desk describes the structure of the relationships you paste. It is not organizational, legal, security, or engineering advice, and it never recommends action against a specific person.
Read the structure
Which nodes matter and why.
Questions
Do I have to sign in?
No. Parsing the edge list and computing every structural metric - degree, components, cut-vertices, bridges, cycles, clustering coefficient, closeness and betweenness centrality, diameter - runs entirely in this browser tab, for free, with no account. Only the AI pass that explains the structure or writes a decision brief is metered, and the bundled example replays a saved run for free.
What counts as a cut-vertex, and why does it matter?
A cut-vertex (articulation point) is a node whose removal splits the graph into more separate pieces than it already had. In an org chart that is a single point of institutional knowledge; in a dependency graph it is a module whose failure fragments the build; in a supply chain it is a single-vendor risk. Graph Desk computes cut-vertices exactly with Tarjan's algorithm, not by estimation.
How is the AI pass kept honest about the graph's structure?
The browser computes every number first and sends those facts into the run as prescan data the model must reconcile against, one entry per flag raised. Every node name the model cites is checked back against the graph you actually pasted, and a name that does not appear in it is shown, marked, rather than silently trusted.
What are the two lanes?
Read the structure explains which nodes matter and why - hubs, cut-vertices, bridge endpoints, sources and sinks - and whether the paste itself looks right. Write the decision brief takes that same graph, and the structural read if you ran it first, and turns it into a short brief for a stated purpose: who to back up, which dependency to break, which single point of failure to fix first.
Is this social engineering, surveillance, or organizational advice?
No. Graph Desk describes the structure of the relationships you paste - it does not recommend firing, restructuring, or monitoring anyone, and its prompt explicitly forbids that framing. Treat every finding as a prompt to verify with the people who actually run the system, not as a decision already made.
What does a run cost, and what if my balance is low?
A worst-case amount is reserved before you run, and only what the run actually uses is charged. The estimate is checked against your balance before the button enables: signed out, it explains that signing in comes first; under the model's minimum, it names the exact shortfall; and a reply cut short by a low balance says so rather than presenting itself as complete.