@henjicc/ai-sdk 0.2.8 → 0.4.0

package/dist/catalog/shared/gptImage25Pricing.js

@@ -0,0 +1,508 @@
+import { GPT25_RATIOS, GPT25_RESOLUTIONS, GPT25_QUALITIES, gpt25Choice, gpt25Count, gpt25Ratio } from './gptImage25.js';
+export const GPT25_OUTPUT_REFERENCE = {
+    "1:1": {
+        "width": 1024,
+        "height": 1024,
+        "tokens": [
+            196,
+            439,
+            1756,
+            3122,
+            7024
+        ]
+    },
+    "3:2": {
+        "width": 1536,
+        "height": 1024,
+        "tokens": [
+            158,
+            343,
+            1372,
+            2459,
+            5488
+        ]
+    },
+    "2:3": {
+        "width": 1024,
+        "height": 1536,
+        "tokens": [
+            158,
+            343,
+            1372,
+            2459,
+            5488
+        ]
+    },
+    "4:3": {
+        "width": 1024,
+        "height": 768,
+        "tokens": [
+            134,
+            301,
+            1204,
+            2140,
+            4815
+        ]
+    },
+    "3:4": {
+        "width": 768,
+        "height": 1024,
+        "tokens": [
+            134,
+            301,
+            1204,
+            2140,
+            4815
+        ]
+    },
+    "5:4": {
+        "width": 1280,
+        "height": 1024,
+        "tokens": [
+            173,
+            378,
+            1510,
+            2702,
+            6119
+        ]
+    },
+    "4:5": {
+        "width": 1024,
+        "height": 1280,
+        "tokens": [
+            173,
+            378,
+            1510,
+            2702,
+            6119
+        ]
+    },
+    "16:9": {
+        "width": 1536,
+        "height": 864,
+        "tokens": [
+            120,
+            280,
+            1078,
+            1917,
+            4312
+        ]
+    },
+    "9:16": {
+        "width": 864,
+        "height": 1536,
+        "tokens": [
+            120,
+            280,
+            1078,
+            1917,
+            4312
+        ]
+    },
+    "2:1": {
+        "width": 2048,
+        "height": 1024,
+        "tokens": [
+            132,
+            295,
+            1180,
+            2098,
+            4720
+        ]
+    },
+    "1:2": {
+        "width": 1024,
+        "height": 2048,
+        "tokens": [
+            132,
+            295,
+            1180,
+            2098,
+            4720
+        ]
+    },
+    "21:9": {
+        "width": 2016,
+        "height": 864,
+        "tokens": [
+            105,
+            225,
+            943,
+            1617,
+            3682
+        ]
+    },
+    "9:21": {
+        "width": 864,
+        "height": 2016,
+        "tokens": [
+            105,
+            225,
+            943,
+            1617,
+            3682
+        ]
+    },
+    "3:1": {
+        "width": 1536,
+        "height": 512,
+        "tokens": [
+            56,
+            134,
+            535,
+            937,
+            2140
+        ]
+    },
+    "1:3": {
+        "width": 512,
+        "height": 1536,
+        "tokens": [
+            56,
+            134,
+            535,
+            937,
+            2140
+        ]
+    },
+    "1:1@2k": {
+        "width": 2048,
+        "height": 2048,
+        "tokens": [
+            397,
+            892,
+            3568,
+            6343,
+            14272
+        ]
+    },
+    "3:2@2k": {
+        "width": 2048,
+        "height": 1360,
+        "tokens": [
+            211,
+            460,
+            1838,
+            3216,
+            7351
+        ]
+    },
+    "2:3@2k": {
+        "width": 1360,
+        "height": 2048,
+        "tokens": [
+            211,
+            460,
+            1838,
+            3216,
+            7351
+        ]
+    },
+    "4:3@2k": {
+        "width": 2048,
+        "height": 1536,
+        "tokens": [
+            247,
+            556,
+            2223,
+            3952,
+            8892
+        ]
+    },
+    "3:4@2k": {
+        "width": 1536,
+        "height": 2048,
+        "tokens": [
+            247,
+            556,
+            2223,
+            3952,
+            8892
+        ]
+    },
+    "5:4@2k": {
+        "width": 2560,
+        "height": 2048,
+        "tokens": [
+            377,
+            826,
+            3303,
+            5911,
+            13385
+        ]
+    },
+    "4:5@2k": {
+        "width": 2048,
+        "height": 2560,
+        "tokens": [
+            377,
+            826,
+            3303,
+            5911,
+            13385
+        ]
+    },
+    "16:9@2k": {
+        "width": 2048,
+        "height": 1152,
+        "tokens": [
+            157,
+            367,
+            1413,
+            2511,
+            5650
+        ]
+    },
+    "9:16@2k": {
+        "width": 1152,
+        "height": 2048,
+        "tokens": [
+            157,
+            367,
+            1413,
+            2511,
+            5650
+        ]
+    },
+    "2:1@2k": {
+        "width": 2688,
+        "height": 1344,
+        "tokens": [
+            180,
+            405,
+            1617,
+            2874,
+            6466
+        ]
+    },
+    "1:2@2k": {
+        "width": 1344,
+        "height": 2688,
+        "tokens": [
+            180,
+            405,
+            1617,
+            2874,
+            6466
+        ]
+    },
+    "21:9@2k": {
+        "width": 2688,
+        "height": 1152,
+        "tokens": [
+            143,
+            306,
+            1285,
+            2202,
+            5016
+        ]
+    },
+    "9:21@2k": {
+        "width": 1152,
+        "height": 2688,
+        "tokens": [
+            143,
+            306,
+            1285,
+            2202,
+            5016
+        ]
+    },
+    "3:1@2k": {
+        "width": 3072,
+        "height": 1024,
+        "tokens": [
+            103,
+            247,
+            988,
+            1729,
+            3952
+        ]
+    },
+    "1:3@2k": {
+        "width": 1024,
+        "height": 3072,
+        "tokens": [
+            103,
+            247,
+            988,
+            1729,
+            3952
+        ]
+    },
+    "1:1@4k": {
+        "width": 2880,
+        "height": 2880,
+        "tokens": [
+            659,
+            1483,
+            5930,
+            10542,
+            23719
+        ]
+    },
+    "3:2@4k": {
+        "width": 3520,
+        "height": 2336,
+        "tokens": [
+            450,
+            982,
+            3926,
+            6870,
+            15703
+        ]
+    },
+    "2:3@4k": {
+        "width": 2336,
+        "height": 3520,
+        "tokens": [
+            450,
+            982,
+            3926,
+            6870,
+            15703
+        ]
+    },
+    "4:3@4k": {
+        "width": 3312,
+        "height": 2480,
+        "tokens": [
+            491,
+            1104,
+            4413,
+            7845,
+            17650
+        ]
+    },
+    "3:4@4k": {
+        "width": 2480,
+        "height": 3312,
+        "tokens": [
+            491,
+            1104,
+            4413,
+            7845,
+            17650
+        ]
+    },
+    "5:4@4k": {
+        "width": 3216,
+        "height": 2576,
+        "tokens": [
+            535,
+            1173,
+            4690,
+            8393,
+            19006
+        ]
+    },
+    "4:5@4k": {
+        "width": 2576,
+        "height": 3216,
+        "tokens": [
+            535,
+            1173,
+            4690,
+            8393,
+            19006
+        ]
+    },
+    "16:9@4k": {
+        "width": 3840,
+        "height": 2160,
+        "tokens": [
+            371,
+            865,
+            3336,
+            5930,
+            13342
+        ]
+    },
+    "9:16@4k": {
+        "width": 2160,
+        "height": 3840,
+        "tokens": [
+            371,
+            865,
+            3336,
+            5930,
+            13342
+        ]
+    },
+    "2:1@4k": {
+        "width": 3840,
+        "height": 1920,
+        "tokens": [
+            300,
+            675,
+            2700,
+            4799,
+            10798
+        ]
+    },
+    "1:2@4k": {
+        "width": 1920,
+        "height": 3840,
+        "tokens": [
+            300,
+            675,
+            2700,
+            4799,
+            10798
+        ]
+    },
+    "21:9@4k": {
+        "width": 3840,
+        "height": 1648,
+        "tokens": [
+            234,
+            500,
+            2099,
+            3598,
+            8196
+        ]
+    },
+    "9:21@4k": {
+        "width": 1648,
+        "height": 3840,
+        "tokens": [
+            234,
+            500,
+            2099,
+            3598,
+            8196
+        ]
+    },
+    "3:1@4k": {
+        "width": 3840,
+        "height": 1280,
+        "tokens": [
+            139,
+            332,
+            1328,
+            2324,
+            5311
+        ]
+    },
+    "1:3@4k": {
+        "width": 1280,
+        "height": 3840,
+        "tokens": [
+            139,
+            332,
+            1328,
+            2324,
+            5311
+        ]
+    }
+};
+export function gpt25SizeReference(params) {
+    const ratio = gpt25Ratio(params, params.gpt25AspectRatio, GPT25_RATIOS);
+    const resolution = gpt25Choice(params.gpt25Resolution, GPT25_RESOLUTIONS, '1K');
+    return GPT25_OUTPUT_REFERENCE[ratio + (resolution === '1K' ? '' : '@' + resolution.toLowerCase())];
+}
+export function gpt25OutputEstimate(params, rate, defaultQuality) {
+    const quality = gpt25Choice(params.gpt25Quality, GPT25_QUALITIES, defaultQuality);
+    const index = ['low', 'medium', 'high', 'xhigh', 'max'].indexOf(quality === 'auto' ? 'max' : quality);
+    return gpt25SizeReference(params).tokens[index] * rate / 1000000 * gpt25Count(params.gpt25Count, 10);
+}

package/dist/llm/defaults.d.ts

@@ -10,7 +10,7 @@ import type { LlmCapabilities, LlmReasoningConfig, LlmReasoningEffort } from './
  */
 export declare const DEFAULT_DEEPSEEK_PROVIDER_ID = "deepseek";
 export declare const DEFAULT_DEEPSEEK_BASE_URL = "https://api.deepseek.com";
-export declare const DEFAULT_DEEPSEEK_MODEL_ID = "deepseek-v4-flash";
+export declare const DEFAULT_DEEPSEEK_MODEL_ID = "deepseek-flash";
 export declare const DEFAULT_PPIO_PROVIDER_ID = "ppio";
 export declare const DEFAULT_PPIO_BASE_URL = "https://api.ppio.com/openai";
 export declare const DEFAULT_PPIO_MODEL_ID = "deepseek/deepseek-v4-flash";

package/dist/llm/defaults.js

@@ -10,7 +10,7 @@ import { applyLlmModelCatalogEntry, findLlmModelCatalogEntry } from './modelCata
  */
 export const DEFAULT_DEEPSEEK_PROVIDER_ID = 'deepseek';
 export const DEFAULT_DEEPSEEK_BASE_URL = 'https://api.deepseek.com';
-export const DEFAULT_DEEPSEEK_MODEL_ID = 'deepseek-v4-flash';
+export const DEFAULT_DEEPSEEK_MODEL_ID = 'deepseek-flash';
 export const DEFAULT_PPIO_PROVIDER_ID = 'ppio';
 export const DEFAULT_PPIO_BASE_URL = 'https://api.ppio.com/openai';
 export const DEFAULT_PPIO_MODEL_ID = 'deepseek/deepseek-v4-flash';

package/dist/llm/discovery.d.ts

@@ -28,6 +28,8 @@ interface RawDiscoveredModel {
     active?: unknown;
 }
 export interface DiscoverModelsOptions {
+    /** 硅基流动的服务端模型分类；其他供应商不接受此选项。 */
+    modelType?: 'chat' | 'embedding' | 'reranker';
     providerFamilyId?: string;
     endpointProfile?: string;
     credentialId?: string;

package/dist/llm/discovery.js

@@ -14,6 +14,8 @@ function optionalPositiveInteger(...values) {
     return null;
 }
 export function parseDiscoveredModel(item) {
+    if (!item || typeof item !== 'object' || Array.isArray(item))
+        return null;
     const modelId = optionalText(item.id);
     if (!modelId)
         return null;
@@ -32,11 +34,18 @@ export async function discoverModels(providerId, baseUrl, runtime, options = {})
         credentialId: options.credentialId,
         baseUrl,
     });
-    const apiKey = await runtime.credentials.get('llm', identity.credentialId);
-    if (options.requireCredential && !apiKey) {
+    const siliconflow = identity.providerFamilyId === 'siliconflow';
+    if (options.modelType !== undefined && (!siliconflow || !['chat', 'embedding', 'reranker'].includes(options.modelType))) {
+        throw new Error('modelType 仅支持硅基流动 chat / embedding / reranker');
+    }
+    const modelType = options.modelType ?? 'chat';
+    const scope = siliconflow && modelType !== 'chat' ? (modelType === 'reranker' ? 'rerank' : 'embedding') : 'llm';
+    const apiKey = await runtime.credentials.get(scope, identity.credentialId);
+    if ((options.requireCredential || siliconflow) && !apiKey) {
         throw new Error(`[api_key_missing] LLM provider "${providerId}" API key is not configured.`);
     }
-    const url = resolveModelsEndpoint(identity.baseUrl ?? baseUrl);
+    const endpoint = resolveModelsEndpoint(identity.baseUrl ?? baseUrl);
+    const url = siliconflow ? `${endpoint}?sub_type=${modelType}` : endpoint;
     const headers = { Accept: 'application/json' };
     if (apiKey) {
         headers.Authorization = `Bearer ${apiKey}`;
@@ -52,8 +61,10 @@ export async function discoverModels(providerId, baseUrl, runtime, options = {})
         if (!response.ok)
             throw new Error(`获取模型列表失败: ${response.status}`);
         const data = await response.json();
-        return (data.data ?? [])
-            .filter(item => !options.activeOnly || item.active !== false)
+        if (!data || !Array.isArray(data.data))
+            throw new Error('获取模型列表失败: 无效的 data 数组');
+        return data.data
+            .filter(item => !options.activeOnly || item?.active !== false)
             .map(parseDiscoveredModel)
             .filter((item) => item !== null);
     }

package/dist/llm/modelCatalogEntries.js

@@ -12,6 +12,22 @@
  * 否则用户会得到"勾了但发过去没反应"的静默失效。
  */
 export const LLM_MODEL_CATALOG_ENTRIES = [
+    // 硅基流动模型页与 Chat VLM 契约；未确认的视频字段和输出上限不猜测。
+    ...[
+        { id: 'kimi-k2.7-code', displayName: 'Kimi K2.7 Code', vendor: 'Moonshot' },
+        { id: 'qwen3.8-27b', displayName: 'Qwen3.8-27B', vendor: 'Qwen' },
+    ].map(model => ({
+        ...model,
+        input: { image: true, video: false, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'json',
+        reasoning: true,
+        sampling: true,
+        contextWindow: 262_144,
+        maxOutputTokens: null,
+        docs: 'docs/model-adaptation/供应商/硅基流动.md',
+    })),
     // ---------------- OpenAI GPT-OSS（Groq 托管） ----------------
     {
         id: 'gpt-oss-20b',
@@ -29,6 +45,22 @@ export const LLM_MODEL_CATALOG_ENTRIES = [
         docs: 'docs/model-adaptation/GPT-OSS-20B/GPT-OSS-20B_Groq.md',
     },
     // ---------------- DeepSeek ----------------
+    {
+        id: 'deepseek-flash',
+        displayName: 'DeepSeek V4.1 Flash',
+        vendor: 'DeepSeek',
+        input: { image: true, video: false, audio: false },
+        toolCall: true,
+        parallelTools: false,
+        structuredOutputMode: 'json',
+        reasoning: true,
+        sampling: true,
+        contextWindow: 1_000_000,
+        maxOutputTokens: 384_000,
+        apiProtocols: ['openai-responses', 'openai-compatible'],
+        note: '支持图片理解；思考模式下 temperature 无效、top_p 下限为 0.95，非思考模式 top_p 固定为 1。',
+        docs: 'docs/llm-adaptation/供应商/DeepSeek.md',
+    },
     {
         id: 'deepseek-v4-flash',
         displayName: 'DeepSeek V4 Flash',

package/dist/llm/providerPresets.d.ts

@@ -1,5 +1,5 @@
 import type { LlmApiProtocol } from './providerProtocol.js';
-import type { LlmModelConfig, LlmProviderConfig, LlmReasoningEffort } from './types.js';
+import type { LlmCapabilities, LlmModelConfig, LlmProviderConfig, LlmReasoningEffort } from './types.js';
 /**
  * 内置供应商预设。
  *
@@ -25,6 +25,8 @@ export interface LlmProviderPreset {
     reasoningConfigurable: boolean;
     /** 添加该供应商时一并建好的推荐模型，能力按内置目录自动标注 */
     modelIds: readonly string[];
+    /** 托管供应商已核验的能力覆盖；不能把原厂能力无条件套给聚合接口。 */
+    modelCapabilities?: Readonly<Record<string, Partial<LlmCapabilities>>>;
     /** 这家供应商已确认可走 Responses 的具体模型；未列出的模型继续走 Chat Completions。 */
     responsesModelIds?: readonly string[];
     websiteUrl: string;

package/dist/llm/providerPresets.js

@@ -1,6 +1,7 @@
 import { DEFAULT_DEEPSEEK_BASE_URL, DEFAULT_DEEPSEEK_PROVIDER_ID, DEFAULT_PPIO_BASE_URL, DEFAULT_PPIO_PROVIDER_ID, createLlmCapabilitiesForModel, } from './defaults.js';
 import { findLlmModelCatalogEntry, modelSupportsLlmApiProtocol } from './modelCatalog.js';
 import { GROQ_PROVIDER_PRESET } from './groq/preset.js';
+import { SILICONFLOW_PROVIDER_PRESET } from './siliconflow/preset.js';
 import { createBigmodelProvider } from './bigmodel/preset.js';
 import { BIGMODEL_ENDPOINT_PROFILE_FAMILY } from './bigmodel/profiles.js';
 import { findProviderMetadata } from '../providers/metadata.js';
@@ -12,6 +13,7 @@ function providerMetadata(providerId) {
 }
 export const LLM_PROVIDER_PRESETS = [
     GROQ_PROVIDER_PRESET,
+    SILICONFLOW_PROVIDER_PRESET,
     {
         providerId: DEFAULT_PPIO_PROVIDER_ID,
         displayName: '派欧云',
@@ -39,8 +41,8 @@ export const LLM_PROVIDER_PRESETS = [
         baseUrl: DEFAULT_DEEPSEEK_BASE_URL,
         reasoning: { enabled: true, effort: 'high' },
         reasoningConfigurable: true,
-        modelIds: ['deepseek-v4-pro', 'deepseek-v4-flash'],
-        responsesModelIds: ['deepseek-v4-pro', 'deepseek-v4-flash', 'deepseek-v4-flash-vision-exp'],
+        modelIds: ['deepseek-flash'],
+        responsesModelIds: ['deepseek-flash', 'deepseek-v4-pro', 'deepseek-v4-flash', 'deepseek-v4-flash-vision-exp'],
         websiteUrl: providerMetadata(DEFAULT_DEEPSEEK_PROVIDER_ID).websiteUrl,
         apiKeyUrl: providerMetadata(DEFAULT_DEEPSEEK_PROVIDER_ID).apiKeyUrl,
         docs: 'docs/llm-adaptation/供应商/DeepSeek.md',
@@ -165,6 +167,7 @@ export function createProviderFromPreset(preset, options = {}) {
     }
     return {
         providerId: options.providerId ?? preset.providerId,
+        ...(preset.providerId === 'siliconflow' ? { providerFamilyId: preset.providerId } : {}),
         credentialId: options.providerId ?? preset.providerId,
         setup: {
             kind: 'preset',
@@ -200,7 +203,7 @@ export function createModelsFromPreset(preset, provider) {
             adapter: provider.adapter,
             apiProtocol: resolvePresetModelApiProtocol(preset, modelId, provider.endpointProfile),
             baseUrl: provider.baseUrl,
-            capabilities: createLlmCapabilitiesForModel(modelId),
+            capabilities: { ...createLlmCapabilitiesForModel(modelId), ...preset.modelCapabilities?.[modelId] },
             catalogId: entry?.id,
             enabled: true,
         };