The AI Generation stages
Prompt, Plan, Generate, Outline, Checks, Connections, Analyze, Accept — what each stage of the AI Generation view does.
The AI Generation view in the editor walks a description through eight stages, shown as subtabs with a progress meter each. A stage you cannot reach yet is disabled, and its tooltip says why. Plan, Generate, Checks, and Connections apply only to multi-section prompts; a short one-shot prompt skips them.
- Prompt
- Describe the graph, pick a provider, and press Generate outline.
- Plan
- The sections the prompt was split into, in generation order.
- Generate
- Each section generated, checked, and accepted in turn, live.
- Outline
- The node hierarchy. Click a step, describe an improvement, and press Apply improvement to change just that step.
- Checks
- Checker findings on the assembled graph.
- Connections
- Every audited pair of connected nodes, problems first.
- Analyze
- The overall verdict. Continue to accept, or go back to the outline.
- Accept
- The levels the draft will apply, then Accept & apply to canvas. The editor switches to Schematic.
When Prompt is blocked
Generation needs a model, a key, and remaining token budget. If one is missing, the Prompt stage explains which, with links to choose a model, raise your token limit, or upgrade the plan.
Open Models & LLM accessChoose the model and check the token budget generation draws on.Related
- Describe a graph and let the builder draft itAI generationWrite what the workflow should do, generate a draft graph, then edit it like anything you drew yourself.
- Refine a generated graphAI generationIterate with follow-up instructions instead of regenerating from scratch, and keep your edits.
- Which model the AI builder usesAI generationThe model, who pays for it, and the monthly token ceiling all come from Models & LLM access.
More in AI generation · Search all help