Guide: AI-Assisted Authoring Workflow

Use command-based AI assistance inside the graph editor to rewrite dialogue, insert nodes, and extend branches — with a draft preview step before any change is applied.

Authoring loop

Open an existing DialogGraph asset in the graph editor.
Configure DialogAISettings and a provider asset if you have not already.
Create a provider asset under Assets/DialogSystemAIExtension/ProviderAssets, duplicate one from Assets/DialogSystemAIExtension/Samples/ExampleAssets/Provider, or use your own folder.
Select an anchor node in the graph (optional — helps the AI target the right location).
Open the AI tab in the right sidebar and type a free-text instruction.
Review the draft preview that appears before anything is written to the graph.
Click Apply to confirm, or discard the draft. Use Ctrl+Z to undo any applied change.

Example instructions

Rewrite this line so the character sounds more nervous
Add three more dialog lines before the end
Insert a PlaySound action after the selected dialog node
Check the graph for structural problems

What the AI may not produce

The AI layer is validated before any draft reaches the graph. It is not permitted to return GUIDs, node positions, edge definitions, or serialised graph JSON. Any response containing these fields is rejected by the safety validator and a failure notice is shown instead of a preview.

Best practices

select an anchor node before issuing positional commands ("insert after this node")
review all proposed action node IDs before confirming — the AI suggests from the registered action list but does not validate game-side bindings
treat all AI output as a draft for human review, not the final source of truth
for OpenAI-compatible services, prefer the OpenAI provider asset with a root baseUrl plus /v1/chat/completions
if a hosted provider returns prose instead of structured output, try a model with structured-output support or use Ollama locally