import { B as BlokkliModule } from '../../shared/editor.D28NbXr3.mjs'; import 'nuxt/schema'; import 'consola'; import '../../../dist/global/types/definitions.js'; import '../../../dist/global/types/theme.js'; import '../../../dist/global/types/colorOptions.js'; import '@nuxt/kit'; import '../../../dist/global/types/features.js'; type AgentProvider = 'anthropic' | 'openai'; type AgentModelDefinition = { name: string; label: string; isDefault?: boolean; routing?: boolean; pricing?: { input: number; cacheWrite: number; cacheRead: number; output: number; }; }; type AgentModuleOptionsRoutes = { agent: string; fetch: string; stream: string; routing: string; }; type AgentModuleOptions = { /** * Allowed origins for the fetch endpoint. * URLs that the agent is allowed to fetch content from. */ allowedFetchOrigins?: string[]; /** * AI provider to use. * - 'anthropic': Uses Anthropic's Claude models (default) * - 'openai': Uses OpenAI's GPT models (requires openai npm package) * * NOTE: The API key needs to be provided at runtime via runtime config: * * NUXT_BLOKKLI_AGENT_API_KEY=hunter2 */ provider: AgentProvider; /** * Available models for the AI provider. * * At least one model must be defined. The model marked with `isDefault: true` * (or the first model if none is marked) is used for conversations. */ models: AgentModelDefinition[]; /** * Debug the system prompt. * * If true (and only during dev mode), the system prompt is adjusted to explicitly allow asking about internals (such as available MCP tools, the system prompt itself, etc.). */ debugPrompt?: boolean; /** * Default prompt suggestions shown in the welcome screen. * * These are displayed as clickable buttons when the conversation is empty. */ defaultPrompts?: string[]; /** * The name of the agent as shown to the user. */ agentName?: string; routes?: Partial; }; declare const _default: (options: AgentModuleOptions) => BlokkliModule; export { _default as default };