Skip to Content

Workspace

The Networks UI is a three-pane workspace. Layout (panel widths, collapsed accordions) is stored with the saved network.

Open it

HowURL
Workbench Solutions → Network Analysis/networks/
Local Workbench SPAhttp://localhost:5270/networks/
Same host as Docker Workbenchhttp://<workbench>/networks/

Use the Workbench login. Standalone Networks login is only for VITE_NETWORKS_STANDALONE_LOGIN=true (dev).

Chrome

Top bar: ecosystem.Ai branding, Settings, Sign Out. Light/dark follows the Networks theme control.

Left header on a model: Close, Export, Save.

ControlWhat it does
Network NameRequired. Saved configs live in Mongo networks.network_configs.
DescriptionOptional notes for operators.
SavePersist source, mappings, analysis flags, and workspace layout.
ExportDownload ecosystem-network-v1 JSON for backup / import.
CloseLeave the editor (return to the model list).

Left — data and graph

Collapsible sections:

  1. Data Source — Mongo, REST, or Spotify OAuth2. See Data Sources.
  2. Graph parameters — graph type, node id/label, edge source/target/weight, optional extra edge mappings, graph mode (standard / social / knowledge), layout spacing, show scores / link weights.
  3. Analytics writeback — merge or export centrality into Mongo after analysis.
  4. Time-based — time field and bracket (minutes → years) for the canvas timeline.

Display max nodes (default 1 000, max 5 000) is the canvas sample, not the analysis universe. Process full collection tells the job to ignore pipeline $limit / $sample while the server still returns only the display sample.

Center — canvas

After the Generate Graph job succeeds:

  • Force-directed (or other) layout of the display sample
  • Size nodes by a metric (PageRank, degree, …)
  • Color nodes by a metric or Community
  • Search nodes by id/label
  • Zoom / pan
  • Timeline buckets when a time field is configured — click a month (or other bracket) to filter; run temporal algorithms from the timeline

Click a node for the details panel (metrics, neighbors). Isolated hubs and a large “CUSTOMER”-style center node are typical when the mapping uses a shared entity field.

The canvas is a sample. Server analysis can cover a much larger edge set (hard ceiling 2 000 000 edges). The UI states that graphs always build as a server job.

Right — analysis

Analysis & Results is enabled once a graph exists.

  1. Tick centrality measures.
  2. Optionally enable Community Detection (Louvain or Girvan–Newman) and Shortest Path.
  3. Run Analysis — another async job; progress shows on the button.
  4. Explain uses the Workbench LLM (/v1) to narrate hubs, communities, or a path. Run analysis first so Explain has metrics.

Results tables, degree-distribution charts, and NL analytics charts can dock in the lower drawer when those settings are on.

Saved models

The landing list loads configs from GET /api/graphtheory/configs. Import uses the same ecosystem-network-v1 export format. Delete is per unique_key.

Last updated on