AI content generation via AI Power Ups

TL;DR

Inject CmsGenerateEntryContentUseCase (from webiny/api/ai-powerups) and call execute(...). It uses the provider the user configured in AI Power Ups and applies an optional Project, Writer Persona, or Reader Persona, so you never pick models, decrypt API keys, or hardcode prompts. It returns the AI-generated entry values as a JSON string; parse it and take the field(s) you want.

Prefer this over a raw Ai.generateText call whenever the point is "apply the user's configured AI setup", it composes the product with itself and is far less plumbing.

Generate content

import { CmsGenerateEntryContentUseCase } from "webiny/api/ai-powerups";

class MyThing {
  constructor(private generate: CmsGenerateEntryContentUseCase.Interface) {}

  async run(model, entry, ctx) {
    const result = await this.generate.execute({
      modelId: model.modelId,
      prompt: `Write a one-sentence marketing summary for "${entry.values.name}". Fill only the "aiSummary" field.`,
      // Any of these are optional; they map to what the user configured in AI Power Ups:
      projectId: ctx?.projectId, // a bundled context (instructions + files + default personas)
      writerPersonaId: ctx?.writerPersonaId, // tone
      readerPersonaId: ctx?.readerPersonaId // audience
    });
    if (result.isFail()) {
      throw result.error; // e.g. "No AI provider configured. Add a provider in AI Power Ups settings."
    }
    return result.value; // { values, telemetry }
  }
}

Register the dependency: dependencies: [CmsGenerateEntryContentUseCase].

Output shape, important

result.value.values is an entry-shaped object keyed by the model's field ids (the use case is built to fill an entry from its schema). The AI decides which fields it fills, so: