All AI automation ultimately calls Google’s Generative AI endpoint via client.Models.GenerateContentAsync(model: model, contents: contents, config: config) inside GeminiClient.GenerateFromTextAndImageAsync (ActionlyPlugin/src/Helpers/GeminiClient.cs (lines 141-144)). The model string saved in model.txt (e.g., gemini-2.5-flash) selects the concrete API endpoint, and the request bundles the desktop screenshot plus the user’s goal.