zhipu-ai-provider/dist/index.d.mts

Vercel AI SDK Custom Provider for Services from Zhipu (Z.ai)

import { ProviderV3, LanguageModelV3, EmbeddingModelV3, ImageModelV3 } from '@ai-sdk/provider';
import { FetchFunction } from '@ai-sdk/provider-utils';

type ZhipuChatModelId = "glm-4.7" | "glm-4.7-flash" | "glm-4.7-flashx" | "glm-4.6" | "glm-4.6-flash" | "glm-4.6-flashx" | "glm-4.5" | "glm-4.5-flash" | "glm-4.5-flashx" | "glm-4.5-air" | "glm-4.5-airx" | "glm-4-flash-250414" | "glm-4-flashx-250414" | "autoglm-phone" | "glm-4.6v" | "glm-4.6v-flash" | "glm-4.6v-flashx" | "glm-4.1v-thinking-flash" | "glm-4.1v-thinking-flashx" | "glm-4v" | "glm-4v-flash" | "glm-z1-air" | "glm-z1-airx" | "glm-z1-flash" | (string & {});
/**
 * Thinking mode configuration for GLM-4.5+ models.
 * Enables deep reasoning capabilities for complex tasks.
 */
interface ZhipuThinkingConfig {
    /**
     * Enable or disable thinking mode.
     * - "enabled": Model will use deep reasoning before responding
     * - "disabled": Standard response without explicit reasoning
     */
    type: "enabled" | "disabled";
    /**
     * Whether to clear thinking content from previous turns.
     * When true, previous reasoning is not retained in context.
     * @default false
     */
    clearThinking?: boolean;
}
interface ZhipuChatSettings {
    /**
     * The unique ID of the end user, helps the platform intervene in illegal activities, generate illegal or improper information, or other abuse by the end user.
     * ID length requirement: at least 6 characters, up to 128 characters.
     */
    userId?: string;
    /**
     * The unique ID of the request, passed by the user side, must be unique;
     * The platform will generate one by default if not provided by the user side.
     */
    requestId?: string;
    /**
     * When do_sample is true, sampling strategy is enabled, when do_sample is false, the sampling strategy temperature, top_p will not take effect
     */
    doSample?: boolean;
    /**
     * Enable thinking/reasoning mode for GLM-4.5+ models.
     * When enabled, the model will perform deep reasoning before responding,
     * which improves performance on complex tasks like coding and multi-step reasoning.
     *
     * @see https://docs.z.ai/guides/llm/glm-4.7
     */
    thinking?: ZhipuThinkingConfig;
}

type ZhipuEmbeddingModelId = "embedding-2" | "embedding-3" | (string & {});
interface ZhipuEmbeddingSettings {
    /**
     * Override the embedding dimension, defaults to 2048.
     * 256, 512, 1024 or 2048 are recommended for embedding-3.
     */
    dimensions?: number;
}

type ZhipuImageModelId = "glm-image" | "cogview-3-flash" | "cogview-4" | (string & {});

interface ZhipuProvider extends ProviderV3 {
    (modelId: ZhipuChatModelId, settings?: ZhipuChatSettings): LanguageModelV3;
    languageModel(modelId: ZhipuChatModelId, settings?: ZhipuChatSettings): LanguageModelV3;
    chat(modelId: ZhipuChatModelId, settings?: ZhipuChatSettings): LanguageModelV3;
    /** @deprecated Use embedding or embeddingModel instead */
    textEmbeddingModel(modelId: ZhipuEmbeddingModelId, settings?: ZhipuEmbeddingSettings): EmbeddingModelV3;
    embedding(modelId: ZhipuEmbeddingModelId, settings?: ZhipuEmbeddingSettings): EmbeddingModelV3;
    embeddingModel(modelId: ZhipuEmbeddingModelId, settings?: ZhipuEmbeddingSettings): EmbeddingModelV3;
    image(modelId: ZhipuImageModelId): ImageModelV3;
    imageModel(modelId: ZhipuImageModelId): ImageModelV3;
}
interface ZhipuProviderSettings {
    baseURL?: string;
    apiKey?: string;
    headers?: Record<string, string>;
    fetch?: FetchFunction;
}
declare function createZhipu(options?: ZhipuProviderSettings): ZhipuProvider;
declare const zhipu: ZhipuProvider;
declare const zai: ZhipuProvider;

export { type ZhipuProvider, type ZhipuProviderSettings, createZhipu, zai, zhipu };