@xiaozhiclaw/provider-core 0.1.0 → 0.1.2

package/dist/builtin-providers.js

@@ -5,6 +5,34 @@
  * model catalog API. External broad catalogs are intentionally not merged in.
  */
 export const BUILTIN_PROVIDERS = [
+    {
+        id: "tei-bge-m3",
+        name: "BAAI bge-m3 (TEI)",
+        transport: "openai-chat",
+        baseUrl: "http://127.0.0.1:8080",
+        apiKeyEnvVars: [],
+        authType: "none",
+        isAggregator: false,
+        defaultModel: "BAAI/bge-m3",
+        models: [
+            {
+                id: "BAAI/bge-m3",
+                aliases: ["bge-m3"],
+                name: "BAAI bge-m3",
+                contextWindow: 8192,
+                maxOutput: 0,
+                toolCall: false,
+                reasoning: false,
+                vision: false,
+                mediaType: "embedding",
+                mediaCapabilities: {
+                    type: "embedding",
+                    dimensions: 1024,
+                    maxTokens: 8192,
+                },
+            },
+        ],
+    },
     // 鈹€鈹€ Tier 1: Major Chinese providers 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€
     // DeepSeek via Anthropic Messages API 鈥?CC-parity transport for tool_use / thinking blocks.
     // See https://api-docs.deepseek.com/guides/anthropic_api
@@ -55,6 +83,9 @@ export const BUILTIN_PROVIDERS = [
     {
         id: "qwen",
         name: "Alibaba Qwen (DashScope)",
+        group: "qwen",
+        variantKind: "standard",
+        billingChannelKind: "paygo",
         transport: "anthropic-messages",
         baseUrl: "https://dashscope.aliyuncs.com/apps/anthropic",
         apiKeyEnvVars: ["DASHSCOPE_API_KEY", "QWEN_API_KEY"],
@@ -1444,9 +1475,13 @@ export const BUILTIN_PROVIDERS = [
     {
         id: "volcengine",
         name: "Doubao / Volcengine",
+        group: "volcengine",
+        variantKind: "standard",
+        billingChannelKind: "paygo",
+        capabilities: ["thinking", "vision", "media"],
         transport: "volcengine-responses",
         baseUrl: "https://ark.cn-beijing.volces.com/api",
-        apiKeyEnvVars: ["ARK_API_KEY", "DOUBAO_API_KEY"],
+        apiKeyEnvVars: ["VOLCANO_ENGINE_API_KEY", "ARK_API_KEY", "DOUBAO_API_KEY"],
         authType: "bearer",
         isAggregator: false,
         defaultModel: "doubao-seed-2-0-lite-260428",
@@ -1585,9 +1620,9 @@ export const BUILTIN_PROVIDERS = [
             // Supports: text-to-image, image-to-image (single/multi-ref), group generation
             // Docs: https://www.volcengine.com/docs/82379/1824121
             {
-                id: "doubao-seedream-5-0-260128",
-                aliases: ["doubao-seedream-5-0"],
-                name: "Doubao Seedream 5.0",
+                id: "doubao-seedream-5-0-lite-260128",
+                aliases: ["doubao-seedream-5-0-lite"],
+                name: "Doubao Seedream 5.0 Lite",
                 contextWindow: 4096,
                 maxOutput: 1,
                 toolCall: false,
@@ -1681,6 +1716,58 @@ export const BUILTIN_PROVIDERS = [
             },
         ],
     },
+    // Doubao subscription / Plan channel. It is a first-class provider variant so
+    // a Plan key never gets silently mixed with the pay-as-you-go Ark API key.
+    {
+        id: "volcengine-plan",
+        name: "Doubao / Volcengine Plan",
+        group: "volcengine",
+        variantKind: "coding-plan",
+        billingChannelKind: "plan",
+        capabilities: ["thinking", "coding"],
+        transport: "openai-chat",
+        baseUrl: "https://ark.cn-beijing.volces.com/api/coding/v3",
+        apiKeyEnvVars: ["VOLCANO_ENGINE_API_KEY", "ARK_API_KEY", "DOUBAO_API_KEY"],
+        authType: "bearer",
+        isAggregator: false,
+        defaultModel: "ark-code-latest",
+        models: [
+            {
+                id: "ark-code-latest",
+                name: "Ark Code Latest (Plan)",
+                contextWindow: 262_144,
+                maxOutput: 65_536,
+                toolCall: true,
+                reasoning: true,
+                vision: false,
+                costInput: 0,
+                costOutput: 0,
+            },
+            {
+                id: "doubao-seed-code",
+                name: "Doubao Seed Code (Plan)",
+                contextWindow: 262_144,
+                maxOutput: 65_536,
+                toolCall: true,
+                reasoning: true,
+                vision: false,
+                costInput: 0,
+                costOutput: 0,
+            },
+            {
+                id: "doubao-seed-code-preview-251028",
+                aliases: ["doubao-seed-code-preview"],
+                name: "Doubao Seed Code Preview (Plan)",
+                contextWindow: 262_144,
+                maxOutput: 65_536,
+                toolCall: true,
+                reasoning: true,
+                vision: false,
+                costInput: 0,
+                costOutput: 0,
+            },
+        ],
+    },
     // 鈹€鈹€ Tier 2: Major International providers 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€
     {
         id: "openai",
@@ -2153,7 +2240,7 @@ export const BUILTIN_PROVIDERS = [
         id: "openrouter",
         name: "OpenRouter",
         transport: "openai-chat",
-        baseUrl: "https://openrouter.ai/api",
+        baseUrl: "https://openrouter.ai/api/v1",
         apiKeyEnvVars: ["OPENROUTER_API_KEY"],
         authType: "bearer",
         isAggregator: true,
@@ -2187,6 +2274,53 @@ export const BUILTIN_PROVIDERS = [
                 costOutput: 25.0,
                 costCacheRead: 0.5,
             },
+            {
+                id: "openai/gpt-5.5",
+                name: "GPT-5.5 (via OpenRouter)",
+                contextWindow: 1050000,
+                maxOutput: 128000,
+                toolCall: true,
+                reasoning: true,
+                vision: true,
+                costInput: 5.0,
+                costOutput: 30.0,
+                costCacheRead: 0.5,
+            },
+            {
+                id: "openai/gpt-5.4",
+                name: "GPT-5.4 (via OpenRouter)",
+                contextWindow: 1050000,
+                maxOutput: 128000,
+                toolCall: true,
+                reasoning: true,
+                vision: true,
+                costInput: 2.5,
+                costOutput: 15.0,
+                costCacheRead: 0.25,
+            },
+            {
+                id: "google/gemini-3.1-pro-preview",
+                name: "Gemini 3.1 Pro (via OpenRouter)",
+                contextWindow: 1048576,
+                maxOutput: 65536,
+                toolCall: true,
+                reasoning: true,
+                reasoningRequired: true,
+                vision: true,
+                costInput: 2.0,
+                costOutput: 12.0,
+            },
+            {
+                id: "google/gemini-3-flash-preview",
+                name: "Gemini 3 Flash (via OpenRouter)",
+                contextWindow: 1048576,
+                maxOutput: 65536,
+                toolCall: true,
+                reasoning: true,
+                vision: true,
+                costInput: 0.50,
+                costOutput: 3.0,
+            },
             {
                 id: "deepseek/deepseek-v4-flash",
                 name: "DeepSeek V4 Flash (via OpenRouter)",
@@ -2207,6 +2341,9 @@ export const BUILTIN_PROVIDERS = [
     {
         id: "qwen-coding",
         name: "Alibaba Qwen (Coding Plan)",
+        group: "qwen",
+        variantKind: "coding-plan",
+        billingChannelKind: "plan",
         transport: "anthropic-messages",
         baseUrl: "https://coding.dashscope.aliyuncs.com/apps/anthropic",
         apiKeyEnvVars: ["DASHSCOPE_API_KEY", "QWEN_API_KEY"],
@@ -2234,4 +2371,26 @@ export const BUILTIN_PROVIDERS = [
             },
         ],
     },
+    {
+        id: "llmrouter",
+        name: "LLMRouter",
+        transport: "openai-responses",
+        baseUrl: "https://www.qlogicagent.com",
+        apiKeyEnvVars: ["QLOGIC_LLMROUTER_API_KEY", "LLMROUTER_API_KEY"],
+        authType: "bearer",
+        isAggregator: true,
+        defaultModel: "llmrouter-default",
+        capabilities: ["thinking", "tool_stream", "vision", "media"],
+        models: [
+            {
+                id: "llmrouter-default",
+                name: "LLMRouter Default",
+                contextWindow: 1000000,
+                maxOutput: 128000,
+                toolCall: true,
+                reasoning: true,
+                vision: true,
+            },
+        ],
+    },
 ];

package/dist/index.d.ts

@@ -26,6 +26,7 @@ export { VolcengineMediaTransport } from "./transports/volcengine-media.js";
 export { OpenAIMediaTransport } from "./transports/openai-media.js";
 export { MiniMaxMediaTransport } from "./transports/minimax-media.js";
 export { GeminiMediaTransport, type GeminiMediaConfig } from "./transports/gemini-media.js";
+export { OpenAICompatibleEmbeddingMediaTransport, type OpenAICompatibleEmbeddingMediaConfig } from "./transports/openai-compatible-embedding-media.js";
 export { GeminiLyriaRealtimeSession, generateRealtimeMusic } from "./transports/gemini-lyria-realtime.js";
 export type { WeightedPrompt, MusicGenerationConfig, MusicScale, MusicGenerationMode, LyriaRealtimeConfig, LyriaRealtimeSessionOptions, AudioChunk } from "./transports/gemini-lyria-realtime.js";
 export { RealtimeTransport } from "./transports/realtime-transport.js";

package/dist/index.js

@@ -22,6 +22,7 @@ export { VolcengineMediaTransport } from "./transports/volcengine-media.js";
 export { OpenAIMediaTransport } from "./transports/openai-media.js";
 export { MiniMaxMediaTransport } from "./transports/minimax-media.js";
 export { GeminiMediaTransport } from "./transports/gemini-media.js";
+export { OpenAICompatibleEmbeddingMediaTransport } from "./transports/openai-compatible-embedding-media.js";
 // Gemini Lyria RealTime (WebSocket streaming music)
 export { GeminiLyriaRealtimeSession, generateRealtimeMusic } from "./transports/gemini-lyria-realtime.js";
 // OpenAI Realtime (WebSocket bidirectional audio/voice)

package/dist/media-client.js

@@ -14,6 +14,7 @@ import { MiniMaxMediaTransport } from "./transports/minimax-media.js";
 import { GeminiMediaTransport } from "./transports/gemini-media.js";
 import { QwenMediaTransport } from "./transports/qwen-media.js";
 import { ZhipuMediaTransport } from "./transports/zhipu-media.js";
+import { OpenAICompatibleEmbeddingMediaTransport } from "./transports/openai-compatible-embedding-media.js";
 // 鈹€鈹€ MediaClient 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€鈹€鈹€鈹€
 export class MediaClient {
     registry;
@@ -142,11 +143,14 @@ function modelSupportsOperation(caps, requiredOp) {
  */
 function createMediaTransport(providerDef) {
     // Remove trailing /anthropic path for providers that use it for chat
-    const rawUrl = providerDef.baseUrl;
+    const rawUrl = providerDef.id === "tei-bge-m3"
+        ? (process.env.TEI_BGE_M3_BASE_URL || process.env.BGE_M3_BASE_URL || providerDef.baseUrl)
+        : providerDef.baseUrl;
     switch (providerDef.id) {
         case "volcengine":
             return new VolcengineMediaTransport({ baseUrl: rawUrl });
         case "openai":
+        case "llmrouter":
             return new OpenAIMediaTransport({ baseUrl: rawUrl });
         case "minimax":
             return new MiniMaxMediaTransport({ baseUrl: rawUrl });
@@ -167,6 +171,8 @@ function createMediaTransport(providerDef) {
             return new ZhipuMediaTransport({
                 baseUrl: rawUrl.replace(/\/api\/coding\/paas\/v4\/?$/, "/api/paas/v4"),
             });
+        case "tei-bge-m3":
+            return new OpenAICompatibleEmbeddingMediaTransport({ baseUrl: rawUrl });
         default:
             // Provider doesn't support media generation
             return undefined;

package/dist/media-transport.d.ts

@@ -11,7 +11,7 @@
  */
 export type MediaType = "image" | "video" | "music" | "music_realtime" | "tts" | "3d" | "stt" | "embedding" | "video_understanding" | "image_understanding" | "voice_clone" | "rerank" | "document_parsing" | "realtime_audio" | "realtime_video";
 export interface MediaRequest {
-    /** Generation model id, e.g. "doubao-seedream-5-0-260128", "gpt-image-2" */
+    /** Generation model id, e.g. "doubao-seedream-5-0-lite-260128", "gpt-image-2" */
     model: string;
     /** What kind of media to generate */
     mediaType: MediaType;
@@ -117,6 +117,8 @@ export interface MediaRequest {
 export interface MediaResult {
     /** URLs of generated media files */
     mediaUrls: string[];
+    /** Provider actually used */
+    provider?: string;
     /** Model actually used */
     model?: string;
     /** Output dimensions / format info */

package/dist/transports/openai-compatible-embedding-media.d.ts

@@ -0,0 +1,12 @@
+import type { MediaRequest, MediaResult, MediaTransport, MediaType } from "../media-transport.js";
+export interface OpenAICompatibleEmbeddingMediaConfig {
+    baseUrl: string;
+    timeoutMs?: number;
+}
+export declare class OpenAICompatibleEmbeddingMediaTransport implements MediaTransport {
+    readonly supportedTypes: readonly MediaType[];
+    private readonly baseUrl;
+    private readonly timeoutMs;
+    constructor(config: OpenAICompatibleEmbeddingMediaConfig);
+    generate(request: MediaRequest, apiKey: string, signal?: AbortSignal): Promise<MediaResult>;
+}

package/dist/transports/openai-compatible-embedding-media.js

@@ -0,0 +1,51 @@
+export class OpenAICompatibleEmbeddingMediaTransport {
+    supportedTypes = ["embedding"];
+    baseUrl;
+    timeoutMs;
+    constructor(config) {
+        this.baseUrl = config.baseUrl.replace(/\/$/, "");
+        this.timeoutMs = config.timeoutMs ?? 30_000;
+    }
+    async generate(request, apiKey, signal) {
+        const start = Date.now();
+        const text = request.metadata?.input ?? request.text ?? request.prompt;
+        if (!text)
+            throw new Error("OpenAICompatibleEmbeddingMediaTransport: text or prompt is required for embedding");
+        const body = {
+            model: request.model,
+            input: text,
+        };
+        if (request.metadata?.dimensions)
+            body.dimensions = request.metadata.dimensions;
+        const headers = { "Content-Type": "application/json" };
+        if (apiKey)
+            headers.Authorization = `Bearer ${apiKey}`;
+        const res = await fetch(`${this.baseUrl}/v1/embeddings`, {
+            method: "POST",
+            headers,
+            body: JSON.stringify(body),
+            signal: signal ?? AbortSignal.timeout(this.timeoutMs),
+        });
+        if (!res.ok) {
+            const textBody = await res.text().catch(() => "");
+            throw new Error(`OpenAI-compatible embedding error ${res.status}: ${textBody}`);
+        }
+        const data = await res.json();
+        const embeddings = (data.data ?? [])
+            .map(item => item.embedding)
+            .filter((item) => Array.isArray(item));
+        const totalTokens = data.usage?.total_tokens ?? data.usage?.prompt_tokens;
+        return {
+            mediaUrls: [],
+            model: data.model ?? request.model,
+            durationMs: Date.now() - start,
+            billingUnit: totalTokens !== undefined ? "per_token" : undefined,
+            billingQuantity: totalTokens,
+            metadata: {
+                embeddings,
+                dimensions: embeddings[0]?.length ?? 0,
+                usage: data.usage,
+            },
+        };
+    }
+}

package/dist/transports/openai-media.js

@@ -61,9 +61,13 @@ export class OpenAIMediaTransport {
             throw new Error(`OpenAI images API error ${res.status}: ${text}`);
         }
         const data = await res.json();
-        const mediaUrls = (data.data ?? [])
-            .map(d => d.url)
-            .filter((u) => !!u);
+        const mediaUrls = (data.data ?? []).flatMap(d => {
+            if (d.url)
+                return [d.url];
+            if (d.b64_json)
+                return [`data:image/webp;base64,${d.b64_json}`];
+            return [];
+        });
         return {
             mediaUrls,
             model: request.model,

package/dist/transports/volcengine-media.js

@@ -140,9 +140,7 @@ export class VolcengineMediaTransport {
             return this.parseStreamingImage(res.body, request, start);
         }
         const data = await res.json();
-        const mediaUrls = (data.data ?? [])
-            .map(d => d.url)
-            .filter((u) => !!u);
+        const mediaUrls = extractVolcengineImageUrls(data.data ?? data);
         return {
             mediaUrls,
             model: request.model,
@@ -770,6 +768,43 @@ export class VolcengineMediaTransport {
         throw new Error("Volcengine task timed out after polling");
     }
 }
+function extractVolcengineImageUrls(value) {
+    const urls = [];
+    const visit = (node) => {
+        if (!node)
+            return;
+        if (Array.isArray(node)) {
+            for (const item of node)
+                visit(item);
+            return;
+        }
+        if (typeof node !== "object")
+            return;
+        const record = node;
+        for (const key of ["url", "image_url", "output_url"]) {
+            const url = record[key];
+            if (typeof url === "string" && url.trim())
+                urls.push(url);
+        }
+        const b64 = record.b64_json;
+        if (typeof b64 === "string" && b64.trim())
+            urls.push(`data:image/webp;base64,${b64}`);
+        for (const key of ["image_urls", "urls", "output_urls"]) {
+            const value = record[key];
+            if (Array.isArray(value)) {
+                for (const item of value) {
+                    if (typeof item === "string" && item.trim())
+                        urls.push(item);
+                }
+            }
+        }
+        for (const child of Object.values(record)) {
+            visit(child);
+        }
+    };
+    visit(value);
+    return [...new Set(urls)];
+}
 function extractGeneratedTokenUsage(result) {
     const usage = (result.usage ?? result.data?.usage);
     const completionTokens = numericUsage(usage?.completion_tokens)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@xiaozhiclaw/provider-core",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "Shared QLogic LLM provider adaptation layer",
   "type": "module",
   "main": "./dist/index.js",