Read and edit each connected input in the shared card/sidebar field rows, move sources between named fields, reorder multi-source inputs, choose a source output, and route it through Extract Output when needed.
Where: Flowgen → select a node or open its fields; use the card rows, node graph, or right inspector.
Before you start
- Open an editable Flowgen canvas. Use node graph mode when you need to see or draw connection noodles.
Key ideas
- One field model, multiple views: An input is a named field with a typed well and the sources connected to it. The shared field-row system presents those same connections on eligible empty cards and in the right inspector; it does not create a second copy of the graph.
- A field’s source order is meaningful: Where a node accepts multiple sources, the ordered chips represent connection order. Text may concatenate in that order and ordered section/media inputs may preserve that sequence, so reordering is not merely visual.
- A wire carries a selected output: A source node with several compatible outputs can expose which output a connected chip reads. Extract Output is a wire-only pass-through for choosing one output explicitly; it does not run or store a result.
Steps
- Select a node. Read every field label before changing a connection; empty fields remain visible so you can tell where a source belongs.
- In node-graph mode, drag from the source socket toward the destination. You can drop the noodle onto the named field well to target that input directly.
- In Card mode or the inspector, use the field’s add-source affordance to connect an existing node, asset, upload, text source, or supported Brain/Brand pointer.
- To move an existing connected source to another input on the same node, drag its source chip and drop it on the other named field. Check the field’s new chip and the graph connection after the move.
- For a multi-source field, drag a chip within its field row to reorder it. Wait for the row order to settle, then confirm the resulting sequence before running.
- When a source offers more than one compatible output, open the chip’s output control (or right-click the chip where offered) and choose the exact output. The chip label should name the output it now reads.
- If you want this output choice represented as an explicit graph step, add Extract Output, connect the source to Value, select the desired output on that incoming source chip, and use Extracted downstream.
- For Groups, distinguish exposed input fields and their internal relay from outer wires; reveal the required group inputs/outputs and map them to the corresponding child handles.
- Run only after checking the intended field, source order, output label, and any paid-operation quote.
What a field row shows
- The destination field label and accepted data type.
- Any connected source chips and their source/output names.
- An add-source control for compatible inputs.
- Actions to edit a connected Text source, replace supported media, remove a connection, or pan to its source.
- The right Properties dock keeps input wells visible for the selected node, including empty wells; eligible empty cards also show their inputs before a generated result exists.
- The connection source can be reordered within supported multi-source fields or moved to another named destination field.
Card, sidebar, and graph boundaries
- Eligible cards can carry their input rows while their content/result surface is empty; content-bearing cards keep their result-focused surface.
- The selected-node inspector uses the same field-row engine in a panel frame; it shows real graph edges whether graph noodles are visible or Card mode hides them.
- Node graph mode is the view for seeing/editing noodles; Card mode is a compact presentation, not a separate graph.
- A node hidden from the compact canvas is not deleted, and its wires remain in the saved flow.
- Group input and output handles are positioned beyond the frame edge so the corresponding outer noodle can be selected and connected without hitting a child node.
- The @ mention picker is anchored to its active text field so the list is not clipped by the node card that contains it.
- Noodles stay black on light canvas and white on dark canvas; selection adds a visible outline and thicker stroke. Wire color is not a data-type key.
Output selection and Extract Output
- Output labels on chips say which source handle is read.
- Output selection is stored on the incoming connection, not as a generated prop on Extract Output.
- Extract Output has Value input and Extracted output; it is pass-through wiring and has no Run result of its own.
- For MCP, inspect actual source handles first and set the chosen sourceHandle on the connection; use get_node_sockets/describe_flow to verify the resulting edge.
Tips
- Name Text sources and important media nodes; source chips then make the graph understandable from either view.
- Use connected Text nodes for authored prompt prose and linked Brand fields for reusable identity data. Do not write inline prompt copy into arbitrary node properties.
- After a cross-field move or reorder, inspect both the field chip list and the graph before a run.
Limitations and important notes
- Only type-compatible inputs and outputs can connect; similar labels do not override socket types.
- Some media sources expose one output, so no output picker is shown.
- The exact editing actions vary by field type and node.
Troubleshooting
I cannot see noodles.
Switch from Card mode to node graph mode. The connections still exist when the compact view hides them.
I dragged a chip but it did not move.
Drop it on another named field to move it, or onto another chip in the same multi-source field to reorder it. Single-source chips can move across fields even when there is nothing to reorder.
Extract Output has no Run button or selectable property.
That is intentional. Choose the source output on its incoming connection chip; the node simply passes that output downstream.