Architecture

FlowGraph is intentionally simple: a small Node/Express server that serves a single canvas page plus a large library of plain-ESM node modules. There is no bundler and no build step for the app.

High-level flow

Browser  ──GET /──►  Express (app.js)
                        │  renders views/index.ejs
                        │  injecting <script type="module"> tags for every
                        │  node discovered by src/module_importer.js
                        ▼
              litegraph canvas (public/js/code.js bootstraps the editor)
                        │
        toolbar + nodes + extensions run entirely client-side
                        │
   optional ──POST /upload_json──►  writes ProjectParameters.json
   optional ──GET  /run_simulation──►  spawns python MainKratos.py (streamed)

Key files

Path	Role
app.js	Express server. Serves public/, renders index.ejs, exposes /upload_json and /run_simulation.
bin/kratos-flowgraph.js	CLI entry point. chdirs to the package root, then launches app.js.
src/module_importer.js	Walks public/js/nodes & public/js/widgets and returns the file list injected into the page.
views/index.ejs	The whole UI: toolbar, <canvas class="graphcanvas">, JSON side panel.
public/js/code.js	Creates the litegraph Editor, exposes window.graph/window.graphcanvas, wires the toolbar.
public/js/litegraph/litegraph.core.js	Vendored litegraph.js engine.
public/js/extensions/*.js	Customisations of the core: categorized menu, custom drawing, node sizing, removal, selection import, ProjectParameters import.
public/js/nodes/**	The node library (see the Node Reference).
public/js/model_manager.js, problem_manager.js	Track model parts / global problem state across connected nodes.
config/default.json	Runtime configuration (port, Kratos paths).

Extensions layer

The files in public/js/extensions/ are loaded after the core library and before the nodes. They adapt vanilla litegraph to FlowGraph's needs:

File	Responsibility
extended_menu.js	Hierarchical, category-based "Add Node" context menu.
draw_node.js, draw_widget.js	Custom node/widget rendering (glyphs, tooltips, error marks).
compute_size.js	Node auto-sizing with margins.
remove_node.js, remove_widget.js	Node/widget deletion.
custom_node_panel.js	Custom node properties panel.
selection_import.js	Selection export/import support.
load_project_parameters.js	The from-JSON converter.

Data flow inside the graph

Each node implements onExecute(). Clicking Generate calls graph.runStep(), which executes nodes in dependency order: source nodes compute their JSON fragments and push them along the connections, intermediate nodes assemble them, and sink nodes (JSON Viewer, Export case) consume the final result. Slot types constrain which connections are legal, so an invalid case is hard to build by accident.