@xiaozhiclaw/provider-core 0.1.0

package/dist/transports/zhipu-media.d.ts

@@ -0,0 +1,82 @@
+/**
+ * Zhipu (GLM) Media Transport 鈥?CogView (image), CogVideoX (video), TTS, STT, Embedding.
+ *
+ * API reference (docs.bigmodel.cn):
+ *   Image sync:  POST /images/generations          (CogView-4, cogview-3-flash)
+ *   Image async: POST /async/images/generations     (glm-image)
+ *   Video async: POST /videos/generations           (CogVideoX)
+ *   TTS sync:    POST /audio/speech                 (glm-tts, returns audio bytes)
+ *   STT sync:    POST /audio/transcriptions         (glm-asr, multipart/form-data)
+ *   Embedding:   POST /embeddings                   (embedding-3/2)
+ *   Async poll:  GET  /async-result/{id}            (unified poll for all async tasks)
+ *
+ * Base URL: https://open.bigmodel.cn/api/paas/v4
+ * Auth: Authorization: Bearer $ZHIPU_API_KEY
+ */
+import type { AsyncMediaTransport, MediaRequest, MediaResult, MediaType } from "../media-transport.js";
+export interface ZhipuMediaConfig {
+    /** Base URL, e.g. "https://open.bigmodel.cn/api/paas/v4" */
+    baseUrl: string;
+    timeoutMs?: number;
+}
+export declare class ZhipuMediaTransport implements AsyncMediaTransport {
+    readonly supportedTypes: readonly MediaType[];
+    private baseUrl;
+    private timeoutMs;
+    constructor(config: ZhipuMediaConfig);
+    generate(request: MediaRequest, apiKey: string, signal?: AbortSignal): Promise<MediaResult>;
+    private generateImage;
+    /** CogView-4 / cogview-3-flash 鈥?sync, returns URL directly */
+    private generateImageSync;
+    /** glm-image 鈥?async submit + poll */
+    private generateImageAsync;
+    private generateVideo;
+    private generateTTS;
+    private generateSTT;
+    private generateEmbedding;
+    private generateVoiceClone;
+    private generateDocumentParsing;
+    private generateRerank;
+    private postJSON;
+    /**
+     * Unified async result polling 鈥?GET /async-result/{id}
+     * Returns the result object when task_status === "SUCCESS".
+     * Throws on "FAIL" or timeout.
+     */
+    private pollAsyncResult;
+    /**
+     * Query a single task status 鈥?GET /async-result/{id}
+     * Zhipu uses a unified async result endpoint for all task types.
+     */
+    getTaskStatus(taskId: string, apiKey: string, signal?: AbortSignal): Promise<{
+        status: string;
+        task: Record<string, unknown>;
+    }>;
+    /**
+     * List recent tasks 鈥?Zhipu does not have a native list endpoint.
+     * Returns empty since individual task query via getTaskStatus() is the primary API.
+     */
+    listVideoTasks(_apiKey: string, _options?: {
+        after?: string;
+        limit?: number;
+        status?: string;
+    }, _signal?: AbortSignal): Promise<Record<string, unknown>>;
+    /**
+     * Cancel/delete is not supported by Zhipu's async API 鈥?throws informative error.
+     * The /async-result/{id} endpoint is read-only.
+     */
+    deleteVideoTask(_taskId: string, _apiKey: string, _signal?: AbortSignal): Promise<void>;
+    /**
+     * List cloned voices 鈥?GET /voice/
+     * Returns all voice clones for the current user.
+     */
+    listVoices(apiKey: string, signal?: AbortSignal): Promise<Array<{
+        voice_id: string;
+        voice_name: string;
+        status: string;
+    }>>;
+    /**
+     * Delete a cloned voice 鈥?POST /voice/delete
+     */
+    deleteVoice(voiceId: string, apiKey: string, signal?: AbortSignal): Promise<void>;
+}

package/dist/transports/zhipu-media.js

@@ -0,0 +1,522 @@
+/**
+ * Zhipu (GLM) Media Transport 鈥?CogView (image), CogVideoX (video), TTS, STT, Embedding.
+ *
+ * API reference (docs.bigmodel.cn):
+ *   Image sync:  POST /images/generations          (CogView-4, cogview-3-flash)
+ *   Image async: POST /async/images/generations     (glm-image)
+ *   Video async: POST /videos/generations           (CogVideoX)
+ *   TTS sync:    POST /audio/speech                 (glm-tts, returns audio bytes)
+ *   STT sync:    POST /audio/transcriptions         (glm-asr, multipart/form-data)
+ *   Embedding:   POST /embeddings                   (embedding-3/2)
+ *   Async poll:  GET  /async-result/{id}            (unified poll for all async tasks)
+ *
+ * Base URL: https://open.bigmodel.cn/api/paas/v4
+ * Auth: Authorization: Bearer $ZHIPU_API_KEY
+ */
+import { writeFileSync, mkdirSync } from "node:fs";
+import { join } from "node:path";
+import { randomUUID } from "node:crypto";
+import { getUserCacheDir } from "../paths.js";
+const DEFAULT_TIMEOUT_MS = 180_000;
+const POLL_INTERVAL_MS = 3_000;
+const MAX_POLL_MS = 600_000; // 10 min max for video
+// Models that use async image generation (submit + poll)
+const ASYNC_IMAGE_MODELS = new Set(["glm-image"]);
+export class ZhipuMediaTransport {
+    supportedTypes = [
+        "image", "video", "tts", "stt", "embedding",
+        "voice_clone", "document_parsing", "rerank",
+    ];
+    baseUrl;
+    timeoutMs;
+    constructor(config) {
+        this.baseUrl = config.baseUrl.replace(/\/+$/, "");
+        this.timeoutMs = config.timeoutMs ?? DEFAULT_TIMEOUT_MS;
+    }
+    async generate(request, apiKey, signal) {
+        switch (request.mediaType) {
+            case "image":
+                return this.generateImage(request, apiKey, signal);
+            case "video":
+                return this.generateVideo(request, apiKey, signal);
+            case "tts":
+                return this.generateTTS(request, apiKey, signal);
+            case "stt":
+                return this.generateSTT(request, apiKey, signal);
+            case "embedding":
+                return this.generateEmbedding(request, apiKey, signal);
+            case "voice_clone":
+                return this.generateVoiceClone(request, apiKey, signal);
+            case "document_parsing":
+                return this.generateDocumentParsing(request, apiKey, signal);
+            case "rerank":
+                return this.generateRerank(request, apiKey, signal);
+            default:
+                throw new Error(`ZhipuMediaTransport: unsupported mediaType "${request.mediaType}"`);
+        }
+    }
+    // 鈹€鈹€ Image (CogView sync / glm-image async) 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    async generateImage(request, apiKey, signal) {
+        if (ASYNC_IMAGE_MODELS.has(request.model)) {
+            return this.generateImageAsync(request, apiKey, signal);
+        }
+        return this.generateImageSync(request, apiKey, signal);
+    }
+    /** CogView-4 / cogview-3-flash 鈥?sync, returns URL directly */
+    async generateImageSync(request, apiKey, signal) {
+        const start = Date.now();
+        const url = `${this.baseUrl}/images/generations`;
+        const body = {
+            model: request.model,
+            prompt: request.prompt,
+        };
+        if (request.size)
+            body.size = request.size;
+        // CogView-4 supports quality param (standard/hd); cogview-3-flash does not
+        if (request.quality && request.model.startsWith("cogview-4"))
+            body.quality = request.quality;
+        const data = await this.postJSON(url, body, apiKey, signal);
+        const items = data.data ?? [];
+        const mediaUrls = items.map(d => d.url).filter((u) => !!u);
+        return {
+            mediaUrls,
+            model: request.model,
+            size: request.size,
+            durationMs: Date.now() - start,
+        };
+    }
+    /** glm-image 鈥?async submit + poll */
+    async generateImageAsync(request, apiKey, signal) {
+        const start = Date.now();
+        const url = `${this.baseUrl}/async/images/generations`;
+        const body = {
+            model: request.model,
+            prompt: request.prompt,
+        };
+        if (request.size)
+            body.size = request.size;
+        if (request.quality)
+            body.quality = request.quality;
+        const submitData = await this.postJSON(url, body, apiKey, signal);
+        const taskId = submitData.id;
+        if (!taskId)
+            throw new Error("Zhipu async image: no task id in response");
+        const result = await this.pollAsyncResult(taskId, apiKey, signal, request.onProgress);
+        const items = (result.data) ?? [];
+        const mediaUrls = items.map(d => d.url).filter((u) => !!u);
+        return {
+            mediaUrls,
+            model: request.model,
+            size: request.size,
+            durationMs: Date.now() - start,
+        };
+    }
+    // 鈹€鈹€ Video (CogVideoX 鈥?async) 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    async generateVideo(request, apiKey, signal) {
+        const start = Date.now();
+        const url = `${this.baseUrl}/videos/generations`;
+        const body = {
+            model: request.model,
+            prompt: request.prompt,
+        };
+        if (request.imageUrl)
+            body.image_url = request.imageUrl;
+        if (request.size)
+            body.size = request.size;
+        if (request.duration)
+            body.duration = request.duration;
+        const submitData = await this.postJSON(url, body, apiKey, signal);
+        const taskId = submitData.id;
+        if (!taskId)
+            throw new Error("Zhipu video generation: no task id in response");
+        const result = await this.pollAsyncResult(taskId, apiKey, signal, request.onProgress);
+        const videoResult = result.video_result;
+        const mediaUrls = (videoResult ?? []).map(v => v.url).filter((u) => !!u);
+        return {
+            mediaUrls,
+            model: request.model,
+            durationMs: Date.now() - start,
+        };
+    }
+    // 鈹€鈹€ TTS (glm-tts 鈥?sync, binary response) 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    async generateTTS(request, apiKey, signal) {
+        const start = Date.now();
+        const text = request.text || request.prompt;
+        if (!text)
+            throw new Error("ZhipuMediaTransport: text or prompt required for TTS");
+        const url = `${this.baseUrl}/audio/speech`;
+        const body = {
+            model: request.model || "glm-tts",
+            input: text,
+        };
+        if (request.voice)
+            body.voice = request.voice;
+        if (request.speed !== undefined)
+            body.speed = request.speed;
+        if (request.audioFormat)
+            body.response_format = request.audioFormat;
+        const res = await fetch(url, {
+            method: "POST",
+            headers: {
+                "Content-Type": "application/json",
+                Authorization: `Bearer ${apiKey}`,
+            },
+            body: JSON.stringify(body),
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!res.ok) {
+            const errText = await res.text().catch(() => "");
+            throw new Error(`Zhipu TTS API error ${res.status}: ${errText}`);
+        }
+        // Response is raw audio bytes 鈥?save to temp file
+        const audioBuffer = Buffer.from(await res.arrayBuffer());
+        const cacheDir = join(getUserCacheDir(), "tts");
+        mkdirSync(cacheDir, { recursive: true });
+        const fmt = request.audioFormat ?? "wav";
+        const filename = `zhipu-tts-${randomUUID()}.${fmt}`;
+        const filePath = join(cacheDir, filename);
+        writeFileSync(filePath, audioBuffer);
+        return {
+            mediaUrls: [`file://${filePath}`],
+            model: request.model || "glm-tts",
+            durationMs: Date.now() - start,
+            billingUnit: "per_character",
+            billingQuantity: text.length,
+        };
+    }
+    // 鈹€鈹€ STT (glm-asr 鈥?sync, multipart) 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    async generateSTT(request, apiKey, signal) {
+        const start = Date.now();
+        if (!request.audioUrl)
+            throw new Error("ZhipuMediaTransport: audioUrl required for STT");
+        const url = `${this.baseUrl}/audio/transcriptions`;
+        // Fetch audio file from URL, then send as multipart
+        const audioRes = await fetch(request.audioUrl, {
+            headers: { "User-Agent": "qlogicagent/1.0" },
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!audioRes.ok)
+            throw new Error(`Failed to fetch audio from ${request.audioUrl}`);
+        const audioBlob = await audioRes.blob();
+        const formData = new FormData();
+        formData.append("model", request.model || "glm-asr-2512");
+        formData.append("file", audioBlob, "audio.wav");
+        const res = await fetch(url, {
+            method: "POST",
+            headers: {
+                Authorization: `Bearer ${apiKey}`,
+            },
+            body: formData,
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!res.ok) {
+            const errText = await res.text().catch(() => "");
+            throw new Error(`Zhipu STT API error ${res.status}: ${errText}`);
+        }
+        const data = await res.json();
+        return {
+            mediaUrls: [],
+            model: request.model || "glm-asr-2512",
+            durationMs: Date.now() - start,
+            metadata: { transcription: data.text ?? "" },
+        };
+    }
+    // 鈹€鈹€ Embedding 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    async generateEmbedding(request, apiKey, signal) {
+        const start = Date.now();
+        const text = request.text || request.prompt;
+        if (!text)
+            throw new Error("ZhipuMediaTransport: text or prompt required for embedding");
+        const url = `${this.baseUrl}/embeddings`;
+        const body = {
+            model: request.model || "embedding-3",
+            input: text,
+        };
+        const data = await this.postJSON(url, body, apiKey, signal);
+        const embeddings = data.data ?? [];
+        return {
+            mediaUrls: [],
+            model: request.model || "embedding-3",
+            durationMs: Date.now() - start,
+            metadata: {
+                embeddings: embeddings.map(e => e.embedding).filter(Boolean),
+            },
+        };
+    }
+    // 鈹€鈹€ Voice Clone (glm-tts-clone 鈥?sync) 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    // POST /voice/clone with audio sample file
+    // See: zhipu-ProviderMax.md 搂20 Voice Clone
+    async generateVoiceClone(request, apiKey, signal) {
+        const start = Date.now();
+        if (!request.audioUrl)
+            throw new Error("ZhipuMediaTransport: audioUrl required for voice_clone (sample audio)");
+        const input = request.text || request.prompt;
+        if (!input)
+            throw new Error("ZhipuMediaTransport: text or prompt required for voice_clone preview input");
+        // Fetch the sample audio
+        const audioRes = await fetch(request.audioUrl, {
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!audioRes.ok)
+            throw new Error(`Failed to fetch audio sample from ${request.audioUrl}`);
+        const audioBlob = await audioRes.blob();
+        const uploadForm = new FormData();
+        uploadForm.append("purpose", "voice-clone-input");
+        uploadForm.append("file", audioBlob, "sample.wav");
+        const uploadRes = await fetch(`${this.baseUrl}/files`, {
+            method: "POST",
+            headers: { Authorization: `Bearer ${apiKey}` },
+            body: uploadForm,
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!uploadRes.ok) {
+            const errText = await uploadRes.text().catch(() => "");
+            throw new Error(`Zhipu Voice Clone file upload error ${uploadRes.status}: ${errText}`);
+        }
+        const uploadData = await uploadRes.json();
+        if (uploadData.error) {
+            throw new Error(`Zhipu Voice Clone file upload rejected: ${uploadData.error.code ?? ""}: ${uploadData.error.message ?? ""}`);
+        }
+        if (!uploadData.id)
+            throw new Error("Zhipu Voice Clone file upload: no file id in response");
+        const body = {
+            model: request.model || "glm-tts-clone",
+            voice_name: String(request.metadata?.voiceName ?? `qlogic_clone_${Date.now()}`),
+            input,
+            file_id: uploadData.id,
+            text: typeof request.metadata?.sampleText === "string" ? request.metadata.sampleText : undefined,
+        };
+        const data = await this.postJSON(`${this.baseUrl}/voice/clone`, body, apiKey, signal);
+        if (data.error) {
+            throw new Error(`Zhipu Voice Clone rejected: ${data.error.code ?? ""}: ${data.error.message ?? ""}`);
+        }
+        return {
+            mediaUrls: [],
+            model: request.model || "glm-tts-clone",
+            durationMs: Date.now() - start,
+            metadata: {
+                voiceId: data.voice,
+                previewFileId: data.file_id,
+                previewFilePurpose: data.file_purpose,
+                requestId: data.request_id,
+                sampleFileId: uploadData.id,
+            },
+        };
+    }
+    // 鈹€鈹€ Document Parsing / OCR (glm-ocr 鈥?sync) 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    // POST /layout_parsing  鈥?accepts file_url or base64
+    // Returns structured page-level parsing results
+    // See: zhipu-ProviderMax.md 搂22 Layout Parsing
+    async generateDocumentParsing(request, apiKey, signal) {
+        const start = Date.now();
+        const file = request.imageUrl || request.audioUrl || request.prompt;
+        if (!file) {
+            throw new Error("ZhipuMediaTransport: imageUrl/audioUrl or prompt (data URL) required for document_parsing");
+        }
+        const url = `${this.baseUrl}/layout_parsing`;
+        const body = {
+            model: request.model || "glm-ocr",
+            file: normalizeZhipuLayoutFile(file, request.metadata),
+        };
+        if (request.metadata?.returnCropImages !== undefined)
+            body.return_crop_images = request.metadata.returnCropImages;
+        if (request.metadata?.needLayoutVisualization !== undefined)
+            body.need_layout_visualization = request.metadata.needLayoutVisualization;
+        if (request.metadata?.startPageId !== undefined)
+            body.start_page_id = request.metadata.startPageId;
+        if (request.metadata?.endPageId !== undefined)
+            body.end_page_id = request.metadata.endPageId;
+        const data = await this.postJSON(url, body, apiKey, signal);
+        const result = data;
+        return {
+            mediaUrls: [],
+            model: request.model || "glm-ocr",
+            durationMs: Date.now() - start,
+            metadata: {
+                content: result.md_results ?? "",
+                mdResults: result.md_results ?? "",
+                layoutDetails: result.layout_details,
+                layoutVisualization: result.layout_visualization,
+                dataInfo: result.data_info,
+                pages: result.data_info?.pages ?? [],
+                pageCount: result.data_info?.num_pages ?? result.data_info?.pages?.length ?? 0,
+                usage: result.usage,
+                requestId: result.request_id,
+            },
+        };
+    }
+    // 鈹€鈹€ Rerank 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    // POST /rerank  鈥?reranks documents by relevance to a query
+    // See: zhipu-ProviderMax.md 搂18 Rerank
+    async generateRerank(request, apiKey, signal) {
+        const start = Date.now();
+        const query = request.prompt;
+        if (!query)
+            throw new Error("ZhipuMediaTransport: prompt (query) required for rerank");
+        if (!request.metadata?.documents) {
+            throw new Error("ZhipuMediaTransport: metadata.documents required for rerank");
+        }
+        const url = `${this.baseUrl}/rerank`;
+        const body = {
+            model: request.model || "rerank",
+            query,
+            documents: request.metadata.documents,
+        };
+        if (request.metadata.topN !== undefined)
+            body.top_n = request.metadata.topN;
+        const data = await this.postJSON(url, body, apiKey, signal);
+        return {
+            mediaUrls: [],
+            model: request.model || "rerank",
+            durationMs: Date.now() - start,
+            metadata: {
+                results: (data.results ?? []).map(r => ({
+                    index: r.index,
+                    relevanceScore: r.relevance_score,
+                    document: r.document,
+                })),
+            },
+        };
+    }
+    // 鈹€鈹€ Shared helpers 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    async postJSON(url, body, apiKey, signal) {
+        const res = await fetch(url, {
+            method: "POST",
+            headers: {
+                "Content-Type": "application/json",
+                Authorization: `Bearer ${apiKey}`,
+            },
+            body: JSON.stringify(body),
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!res.ok) {
+            const errText = await res.text().catch(() => "");
+            throw new Error(`Zhipu API error ${res.status}: ${errText}`);
+        }
+        return res.json();
+    }
+    /**
+     * Unified async result polling 鈥?GET /async-result/{id}
+     * Returns the result object when task_status === "SUCCESS".
+     * Throws on "FAIL" or timeout.
+     */
+    async pollAsyncResult(taskId, apiKey, signal, onProgress) {
+        const deadline = Date.now() + MAX_POLL_MS;
+        const pollUrl = `${this.baseUrl}/async-result/${encodeURIComponent(taskId)}`;
+        while (Date.now() < deadline) {
+            signal?.throwIfAborted();
+            const res = await fetch(pollUrl, {
+                method: "GET",
+                headers: { Authorization: `Bearer ${apiKey}` },
+                signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+            });
+            if (!res.ok) {
+                const errText = await res.text().catch(() => "");
+                throw new Error(`Zhipu async-result error ${res.status}: ${errText}`);
+            }
+            const data = await res.json();
+            if (data.task_status === "SUCCESS") {
+                onProgress?.(100, "completed", taskId);
+                return data;
+            }
+            if (data.task_status === "FAIL") {
+                throw new Error("Zhipu async task failed");
+            }
+            // PROCESSING 鈥?report progress and wait
+            const elapsed = Date.now() - (deadline - MAX_POLL_MS);
+            onProgress?.(Math.min(95, Math.round((elapsed / MAX_POLL_MS) * 100)), data.task_status ?? "running", taskId);
+            await new Promise(r => setTimeout(r, POLL_INTERVAL_MS));
+        }
+        throw new Error(`Zhipu async task timed out after ${MAX_POLL_MS / 1000}s`);
+    }
+    // 鈹€鈹€ AsyncMediaTransport: Task Management 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    /**
+     * Query a single task status 鈥?GET /async-result/{id}
+     * Zhipu uses a unified async result endpoint for all task types.
+     */
+    async getTaskStatus(taskId, apiKey, signal) {
+        const url = `${this.baseUrl}/async-result/${encodeURIComponent(taskId)}`;
+        const res = await fetch(url, {
+            method: "GET",
+            headers: { Authorization: `Bearer ${apiKey}` },
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!res.ok) {
+            const errText = await res.text().catch(() => "");
+            throw new Error(`Zhipu task query error ${res.status}: ${errText}`);
+        }
+        const data = await res.json();
+        const taskStatus = data.task_status ?? "unknown";
+        // Normalize Zhipu statuses (PROCESSING/SUCCESS/FAIL) to lower-case
+        const normalizedStatus = taskStatus === "SUCCESS" ? "succeeded"
+            : taskStatus === "FAIL" ? "failed"
+                : taskStatus === "PROCESSING" ? "running"
+                    : taskStatus.toLowerCase();
+        return { status: normalizedStatus, task: data };
+    }
+    /**
+     * List recent tasks 鈥?Zhipu does not have a native list endpoint.
+     * Returns empty since individual task query via getTaskStatus() is the primary API.
+     */
+    async listVideoTasks(_apiKey, _options, _signal) {
+        // Zhipu has no list endpoint; callers should use getTaskStatus() for direct lookup
+        return { data: [] };
+    }
+    /**
+     * Cancel/delete is not supported by Zhipu's async API 鈥?throws informative error.
+     * The /async-result/{id} endpoint is read-only.
+     */
+    async deleteVideoTask(_taskId, _apiKey, _signal) {
+        throw new Error("Zhipu does not support task cancellation. Tasks complete or timeout automatically.");
+    }
+    // 鈹€鈹€ T17: Voice Management API 鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€鈹€
+    /**
+     * List cloned voices 鈥?GET /voice/
+     * Returns all voice clones for the current user.
+     */
+    async listVoices(apiKey, signal) {
+        const url = `${this.baseUrl}/voice/list`;
+        const res = await fetch(url, {
+            method: "GET",
+            headers: { Authorization: `Bearer ${apiKey}` },
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!res.ok) {
+            const errText = await res.text().catch(() => "");
+            throw new Error(`Zhipu voice list error ${res.status}: ${errText}`);
+        }
+        const data = await res.json();
+        return (data.voice_list ?? []).map(voice => ({
+            voice_id: voice.voice_id ?? voice.voice ?? "",
+            voice_name: voice.voice_name ?? "",
+            status: voice.status ?? "",
+        }));
+    }
+    /**
+     * Delete a cloned voice 鈥?POST /voice/delete
+     */
+    async deleteVoice(voiceId, apiKey, signal) {
+        const url = `${this.baseUrl}/voice/delete`;
+        const res = await fetch(url, {
+            method: "POST",
+            headers: {
+                "Content-Type": "application/json",
+                Authorization: `Bearer ${apiKey}`,
+            },
+            body: JSON.stringify({ voice: voiceId }),
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!res.ok) {
+            const errText = await res.text().catch(() => "");
+            throw new Error(`Zhipu voice delete error ${res.status}: ${errText}`);
+        }
+    }
+}
+function normalizeZhipuLayoutFile(file, metadata) {
+    if (/^(https?:|data:)/i.test(file))
+        return file;
+    const mimeType = typeof metadata?.mimeType === "string"
+        ? metadata.mimeType
+        : "application/pdf";
+    return `data:${mimeType};base64,${file}`;
+}

package/dist/transports/zhipu-tool-api.d.ts

@@ -0,0 +1,35 @@
+/**
+ * ZhipuToolAPI 鈥?Zhipu-specific utility endpoints.
+ *
+ * Implements ProviderToolAPI for Zhipu GLM platform independent APIs:
+ *   C1: Web Search  鈥?POST /tools/web-search
+ *   C2: Reader      鈥?POST /tools/reader (extract web page content)
+ *   C3: Tokenizer   鈥?POST /tokenizer
+ *   C4: Moderations 鈥?POST /moderations
+ *
+ * Base URL: https://open.bigmodel.cn/api/paas/v4
+ * Auth: Authorization: Bearer $ZHIPU_API_KEY
+ *
+ * C5 (File Parser) is handled by document_parsing media handler.
+ * C6 (Realtime API) requires WebSocket 鈥?out of scope for this interface.
+ */
+import type { ProviderToolAPI, ProviderToolCapability, WebSearchResult, ReaderResult, TokenizerResult, ModerationResult } from "../provider-tool-api.js";
+export interface ZhipuToolAPIConfig {
+    baseUrl: string;
+    apiKey: string;
+    timeoutMs?: number;
+}
+export declare class ZhipuToolAPI implements ProviderToolAPI {
+    readonly capabilities: readonly ProviderToolCapability[];
+    private baseUrl;
+    private apiKey;
+    private timeoutMs;
+    constructor(config: ZhipuToolAPIConfig);
+    webSearch(query: string, options?: {
+        maxResults?: number;
+    }): Promise<WebSearchResult[]>;
+    reader(pageUrl: string): Promise<ReaderResult>;
+    tokenize(text: string, model: string): Promise<TokenizerResult>;
+    moderate(text: string): Promise<ModerationResult>;
+    private postJSON;
+}