init

2025-05-25 14:01:37 +08:00
commit 8f52a10ae0
42 changed files with 1946 additions and 0 deletions
--- a/src/provider/chat-adapter/custom.ts
+++ b/src/provider/chat-adapter/custom.ts
@@ -0,0 +1,14 @@
+import { BaseChat, BaseChatOptions } from '../core/chat.ts';
+
+export type OllamaOptions = BaseChatOptions;
+
+/**
+ * 自定义模型
+ */
+export class Custom extends BaseChat {
+  static BASE_URL = 'https://api.deepseek.com/v1/';
+  constructor(options: OllamaOptions) {
+    const baseURL = options.baseURL || Custom.BASE_URL;
+    super({ ...(options as BaseChatOptions), baseURL: baseURL });
+  }
+}
--- a/src/provider/chat-adapter/deepseek.ts
+++ b/src/provider/chat-adapter/deepseek.ts
@@ -0,0 +1,10 @@
+import { BaseChat, BaseChatOptions } from '../core/chat.ts';
+
+export type DeepSeekOptions = Partial<BaseChatOptions>;
+export class DeepSeek extends BaseChat {
+  static BASE_URL = 'https://api.deepseek.com/v1/';
+  constructor(options: DeepSeekOptions) {
+    const baseURL = options.baseURL || DeepSeek.BASE_URL;
+    super({ ...(options as BaseChatOptions), baseURL: baseURL });
+  }
+}
--- a/src/provider/chat-adapter/model-scope.ts
+++ b/src/provider/chat-adapter/model-scope.ts
@@ -0,0 +1,11 @@
+// https://api-inference.modelscope.cn/v1/
+import { BaseChat, BaseChatOptions } from '../core/chat.ts';
+
+export type ModelScopeOptions = Partial<BaseChatOptions>;
+export class ModelScope extends BaseChat {
+  static BASE_URL = 'https://api-inference.modelscope.cn/v1/';
+  constructor(options: ModelScopeOptions) {
+    const baseURL = options.baseURL || ModelScope.BASE_URL;
+    super({ ...options, baseURL: baseURL } as any);
+  }
+}
--- a/src/provider/chat-adapter/ollama.ts
+++ b/src/provider/chat-adapter/ollama.ts
@@ -0,0 +1,47 @@
+import { BaseChat, BaseChatOptions } from '../core/index.ts';
+import type { ChatMessage, ChatMessageOptions } from '../core/index.ts';
+
+export type OllamaOptions = Partial<BaseChatOptions>;
+
+type OllamaModel = {
+  name: string;
+  model: string;
+  modified_at: string;
+
+  size: number;
+  digest: string;
+  details: {
+    parent_model: string;
+    format: string; // example: gguf
+    family: string; // example qwen
+    families: string[];
+    parameter_size: string;
+    quantization_level: string; // example: Q4_K_M Q4_0
+  };
+};
+export class Ollama extends BaseChat {
+  static BASE_URL = 'http://localhost:11434/v1';
+  constructor(options: OllamaOptions) {
+    const baseURL = options.baseURL || Ollama.BASE_URL;
+    super({ ...(options as BaseChatOptions), baseURL: baseURL });
+  }
+  async chat(messages: ChatMessage[], options?: ChatMessageOptions) {
+    const res = await super.chat(messages, options);
+    console.log('thunk', this.getChatUsage());
+    return res;
+  }
+  /**
+   * 获取模型列表
+   * @returns
+   */
+  async listModels(): Promise<{ models: OllamaModel[] }> {
+    const _url = new URL(this.baseURL);
+    const tagsURL = new URL('/api/tags', _url);
+    return this.openai.get(tagsURL.toString());
+  }
+  async listRunModels(): Promise<{ models: OllamaModel[] }> {
+    const _url = new URL(this.baseURL);
+    const tagsURL = new URL('/api/ps', _url);
+    return this.openai.get(tagsURL.toString());
+  }
+}
--- a/src/provider/chat-adapter/siliconflow.ts
+++ b/src/provider/chat-adapter/siliconflow.ts
@@ -0,0 +1,39 @@
+import { BaseChat, BaseChatOptions } from '../core/chat.ts';
+import { OpenAI } from 'openai';
+
+export type SiliconFlowOptions = Partial<BaseChatOptions>;
+
+type SiliconFlowUsageData = {
+  id: string;
+  name: string;
+  image: string;
+  email: string;
+  isAdmin: boolean;
+  balance: string;
+  status: 'normal' | 'suspended' | 'expired' | string; // 状态
+  introduce: string;
+  role: string;
+  chargeBalance: string;
+  totalBalance: string;
+  category: string;
+};
+type SiliconFlowUsageResponse = {
+  code: number;
+  message: string;
+  status: boolean;
+  data: SiliconFlowUsageData;
+};
+export class SiliconFlow extends BaseChat {
+  static BASE_URL = 'https://api.siliconflow.cn/v1';
+  constructor(options: SiliconFlowOptions) {
+    const baseURL = options.baseURL || SiliconFlow.BASE_URL;
+    super({ ...(options as BaseChatOptions), baseURL: baseURL });
+  }
+  async getUsageInfo(): Promise<SiliconFlowUsageResponse> {
+    return this.openai.get('/user/info');
+  }
+  async chat(messages: OpenAI.Chat.Completions.ChatCompletionMessageParam[], options?: Partial<OpenAI.Chat.Completions.ChatCompletionCreateParams>) {
+    const res = await super.chat(messages, options);
+    return res;
+  }
+}
--- a/src/provider/chat-adapter/volces.ts
+++ b/src/provider/chat-adapter/volces.ts
@@ -0,0 +1,10 @@
+import { BaseChat, BaseChatOptions } from '../core/chat.ts';
+
+export type VolcesOptions = Partial<BaseChatOptions>;
+export class Volces extends BaseChat {
+  static BASE_URL = 'https://ark.cn-beijing.volces.com/api/v3/';
+  constructor(options: VolcesOptions) {
+    const baseURL = options.baseURL || Volces.BASE_URL;
+    super({ ...(options as BaseChatOptions), baseURL: baseURL });
+  }
+}
--- a/src/provider/core/chat.ts
+++ b/src/provider/core/chat.ts
@@ -0,0 +1,143 @@
+import { OpenAI } from 'openai';
+import type {
+  BaseChatInterface,
+  ChatMessageComplete,
+  ChatMessage,
+  ChatMessageOptions,
+  BaseChatUsageInterface,
+  ChatStream,
+  EmbeddingMessage,
+  EmbeddingMessageComplete,
+} from './type.ts';
+
+export type BaseChatOptions<T = Record<string, any>> = {
+  /**
+   * 默认baseURL
+   */
+  baseURL: string;
+  /**
+   * 默认模型
+   */
+  model?: string;
+  /**
+   * 默认apiKey
+   */
+  apiKey: string;
+  /**
+   * 是否在浏览器中使用
+   */
+  isBrowser?: boolean;
+  /**
+   * 是否流式输出， 默认 false
+   */
+  stream?: boolean;
+} & T;
+
+export class BaseChat implements BaseChatInterface, BaseChatUsageInterface {
+  /**
+   * 默认baseURL
+   */
+  baseURL: string;
+  /**
+   * 默认模型
+   */
+  model: string;
+  /**
+   * 默认apiKey
+   */
+  apiKey: string;
+  /**
+   * 是否在浏览器中使用
+   */
+  isBrowser: boolean;
+  /**
+   * openai实例
+   */
+  openai: OpenAI;
+
+  prompt_tokens: number;
+  total_tokens: number;
+  completion_tokens: number;
+
+  constructor(options: BaseChatOptions) {
+    this.baseURL = options.baseURL;
+    this.model = options.model;
+    this.apiKey = options.apiKey;
+    this.isBrowser = options.isBrowser ?? false;
+    this.openai = new OpenAI({
+      apiKey: this.apiKey,
+      baseURL: this.baseURL,
+      dangerouslyAllowBrowser: this.isBrowser,
+    });
+  }
+  /**
+   * 聊天
+   */
+  async chat(messages: ChatMessage[], options?: ChatMessageOptions): Promise<ChatMessageComplete> {
+    const createParams: OpenAI.Chat.Completions.ChatCompletionCreateParams = {
+      model: this.model,
+      messages,
+      ...options,
+      stream: false,
+    };
+    const res = (await this.openai.chat.completions.create(createParams)) as ChatMessageComplete;
+    this.prompt_tokens = res.usage?.prompt_tokens ?? 0;
+    this.total_tokens = res.usage?.total_tokens ?? 0;
+    this.completion_tokens = res.usage?.completion_tokens ?? 0;
+    return res;
+  }
+  async chatStream(messages: ChatMessage[], options?: ChatMessageOptions) {
+    const createParams: OpenAI.Chat.Completions.ChatCompletionCreateParams = {
+      model: this.model,
+      messages,
+      ...options,
+      stream: true,
+    };
+    if (createParams.response_format) {
+      throw new Error('response_format is not supported in stream mode');
+    }
+    return this.openai.chat.completions.create(createParams) as unknown as ChatStream;
+  }
+
+  /**
+   * 测试
+   */
+  test() {
+    return this.chat([{ role: 'user', content: 'Hello, world!' }]);
+  }
+  /**
+   * 获取聊天使用情况
+   * @returns
+   */
+  getChatUsage() {
+    return {
+      prompt_tokens: this.prompt_tokens,
+      total_tokens: this.total_tokens,
+      completion_tokens: this.completion_tokens,
+    };
+  }
+  getHeaders(headers?: Record<string, string>) {
+    return {
+      'Content-Type': 'application/json',
+      Authorization: `Bearer ${this.apiKey}`,
+      ...headers,
+    };
+  }
+  /**
+   * 生成embedding 内部
+   * @param text
+   * @returns
+   */
+  async generateEmbeddingCore(text: string | string[], options?: EmbeddingMessage): Promise<EmbeddingMessageComplete> {
+    const embeddingModel = options?.model || this.model;
+    const res = await this.openai.embeddings.create({
+      model: embeddingModel,
+      input: text,
+      encoding_format: 'float',
+      ...options,
+    });
+    this.prompt_tokens += res.usage.prompt_tokens;
+    this.total_tokens += res.usage.total_tokens;
+    return res;
+  }
+}
--- a/src/provider/core/index.ts
+++ b/src/provider/core/index.ts
@@ -0,0 +1,27 @@
+import { ChatStream } from './type.ts';
+
+// export type { BaseChat, BaseChatOptions } from './chat.ts';
+export * from './chat.ts'
+// export {
+//   ChatMessage,
+//   ChatMessageOptions, //
+//   ChatMessageComplete,
+//   ChatMessageStream,
+//   BaseChatInterface,
+//   BaseChatUsageInterface,
+//   ChatStream,
+//   EmbeddingMessage,
+//   EmbeddingMessageComplete,
+// } from './type.ts';
+export * from './type.ts'
+/**
+ *  for await (const chunk of chatStream) {
+ *   console.log(chunk);
+ *  }
+ * @param chatStream 
+ */
+export const readStream = async (chatStream: ChatStream) => {
+  for await (const chunk of chatStream) {
+    console.log(chunk);
+  }
+};
--- a/src/provider/core/text-regex.ts
+++ b/src/provider/core/text-regex.ts
@@ -0,0 +1,105 @@
+// Updated: Aug. 20, 2024
+// Live demo: https://jina.ai/tokenizer
+// LICENSE: Apache-2.0 (https://www.apache.org/licenses/LICENSE-2.0)
+// COPYRIGHT: Jina AI
+
+// Define variables for magic numbers
+const MAX_HEADING_LENGTH = 7;
+const MAX_HEADING_CONTENT_LENGTH = 200;
+const MAX_HEADING_UNDERLINE_LENGTH = 200;
+const MAX_HTML_HEADING_ATTRIBUTES_LENGTH = 100;
+const MAX_LIST_ITEM_LENGTH = 200;
+const MAX_NESTED_LIST_ITEMS = 6;
+const MAX_LIST_INDENT_SPACES = 7;
+const MAX_BLOCKQUOTE_LINE_LENGTH = 200;
+const MAX_BLOCKQUOTE_LINES = 15;
+const MAX_CODE_BLOCK_LENGTH = 1500;
+const MAX_CODE_LANGUAGE_LENGTH = 20;
+const MAX_INDENTED_CODE_LINES = 20;
+const MAX_TABLE_CELL_LENGTH = 200;
+const MAX_TABLE_ROWS = 20;
+const MAX_HTML_TABLE_LENGTH = 2000;
+const MIN_HORIZONTAL_RULE_LENGTH = 3;
+const MAX_SENTENCE_LENGTH = 400;
+const MAX_QUOTED_TEXT_LENGTH = 300;
+const MAX_PARENTHETICAL_CONTENT_LENGTH = 200;
+const MAX_NESTED_PARENTHESES = 5;
+const MAX_MATH_INLINE_LENGTH = 100;
+const MAX_MATH_BLOCK_LENGTH = 500;
+const MAX_PARAGRAPH_LENGTH = 1000;
+const MAX_STANDALONE_LINE_LENGTH = 800;
+const MAX_HTML_TAG_ATTRIBUTES_LENGTH = 100;
+const MAX_HTML_TAG_CONTENT_LENGTH = 1000;
+const LOOKAHEAD_RANGE = 100;  // Number of characters to look ahead for a sentence boundary
+
+const AVOID_AT_START = `[\\s\\]})>,']`;
+const PUNCTUATION = `[.!?…]|\\.{3}|[\\u2026\\u2047-\\u2049]|[\\p{Emoji_Presentation}\\p{Extended_Pictographic}]`;
+const QUOTE_END = `(?:'(?=\`)|''(?=\`\`))`;
+const SENTENCE_END = `(?:${PUNCTUATION}(?<!${AVOID_AT_START}(?=${PUNCTUATION}))|${QUOTE_END})(?=\\S|$)`;
+const SENTENCE_BOUNDARY = `(?:${SENTENCE_END}|(?=[\\r\\n]|$))`;
+const LOOKAHEAD_PATTERN = `(?:(?!${SENTENCE_END}).){1,${LOOKAHEAD_RANGE}}${SENTENCE_END}`;
+const NOT_PUNCTUATION_SPACE = `(?!${PUNCTUATION}\\s)`;
+const SENTENCE_PATTERN = `${NOT_PUNCTUATION_SPACE}(?:[^\\r\\n]{1,{MAX_LENGTH}}${SENTENCE_BOUNDARY}|[^\\r\\n]{1,{MAX_LENGTH}}(?=${PUNCTUATION}|${QUOTE_END})(?:${LOOKAHEAD_PATTERN})?)${AVOID_AT_START}*`;
+
+
+export const textSplitter = new RegExp(
+    "(" +
+    // 1. Headings (Setext-style, Markdown, and HTML-style, with length constraints)
+    `(?:^(?:[#*=-]{1,${MAX_HEADING_LENGTH}}|\\w[^\\r\\n]{0,${MAX_HEADING_CONTENT_LENGTH}}\\r?\\n[-=]{2,${MAX_HEADING_UNDERLINE_LENGTH}}|<h[1-6][^>]{0,${MAX_HTML_HEADING_ATTRIBUTES_LENGTH}}>)[^\\r\\n]{1,${MAX_HEADING_CONTENT_LENGTH}}(?:</h[1-6]>)?(?:\\r?\\n|$))` +
+    "|" +
+    // New pattern for citations
+    `(?:\\[[0-9]+\\][^\\r\\n]{1,${MAX_STANDALONE_LINE_LENGTH}})` +
+    "|" +
+    // 2. List items (bulleted, numbered, lettered, or task lists, including nested, up to three levels, with length constraints)
+    `(?:(?:^|\\r?\\n)[ \\t]{0,3}(?:[-*+•]|\\d{1,3}\\.\\w\\.|\\[[ xX]\\])[ \\t]+${SENTENCE_PATTERN.replace(/{MAX_LENGTH}/g, String(MAX_LIST_ITEM_LENGTH))}` +
+    `(?:(?:\\r?\\n[ \\t]{2,5}(?:[-*+•]|\\d{1,3}\\.\\w\\.|\\[[ xX]\\])[ \\t]+${SENTENCE_PATTERN.replace(/{MAX_LENGTH}/g, String(MAX_LIST_ITEM_LENGTH))}){0,${MAX_NESTED_LIST_ITEMS}}` +
+    `(?:\\r?\\n[ \\t]{4,${MAX_LIST_INDENT_SPACES}}(?:[-*+•]|\\d{1,3}\\.\\w\\.|\\[[ xX]\\])[ \\t]+${SENTENCE_PATTERN.replace(/{MAX_LENGTH}/g, String(MAX_LIST_ITEM_LENGTH))}){0,${MAX_NESTED_LIST_ITEMS}})?)` +
+    "|" +
+    // 3. Block quotes (including nested quotes and citations, up to three levels, with length constraints)
+    `(?:(?:^>(?:>|\\s{2,}){0,2}${SENTENCE_PATTERN.replace(/{MAX_LENGTH}/g, String(MAX_BLOCKQUOTE_LINE_LENGTH))}\\r?\\n?){1,${MAX_BLOCKQUOTE_LINES}})` +
+    "|" +
+    // 4. Code blocks (fenced, indented, or HTML pre/code tags, with length constraints)
+    `(?:(?:^|\\r?\\n)(?:\`\`\`|~~~)(?:\\w{0,${MAX_CODE_LANGUAGE_LENGTH}})?\\r?\\n[\\s\\S]{0,${MAX_CODE_BLOCK_LENGTH}}?(?:\`\`\`|~~~)\\r?\\n?` +
+    `|(?:(?:^|\\r?\\n)(?: {4}|\\t)[^\\r\\n]{0,${MAX_LIST_ITEM_LENGTH}}(?:\\r?\\n(?: {4}|\\t)[^\\r\\n]{0,${MAX_LIST_ITEM_LENGTH}}){0,${MAX_INDENTED_CODE_LINES}}\\r?\\n?)` +
+    `|(?:<pre>(?:<code>)?[\\s\\S]{0,${MAX_CODE_BLOCK_LENGTH}}?(?:</code>)?</pre>))` +
+    "|" +
+    // 5. Tables (Markdown, grid tables, and HTML tables, with length constraints)
+    `(?:(?:^|\\r?\\n)(?:\\|[^\\r\\n]{0,${MAX_TABLE_CELL_LENGTH}}\\|(?:\\r?\\n\\|[-:]{1,${MAX_TABLE_CELL_LENGTH}}\\|){0,1}(?:\\r?\\n\\|[^\\r\\n]{0,${MAX_TABLE_CELL_LENGTH}}\\|){0,${MAX_TABLE_ROWS}}` +
+    `|<table>[\\s\\S]{0,${MAX_HTML_TABLE_LENGTH}}?</table>))` +
+    "|" +
+    // 6. Horizontal rules (Markdown and HTML hr tag)
+    `(?:^(?:[-*_]){${MIN_HORIZONTAL_RULE_LENGTH},}\\s*$|<hr\\s*/?>)` +
+    "|" +
+    // 10. Standalone lines or phrases (including single-line blocks and HTML elements, with length constraints)
+    `(?!${AVOID_AT_START})(?:^(?:<[a-zA-Z][^>]{0,${MAX_HTML_TAG_ATTRIBUTES_LENGTH}}>)?${SENTENCE_PATTERN.replace(/{MAX_LENGTH}/g, String(MAX_STANDALONE_LINE_LENGTH))}(?:</[a-zA-Z]+>)?(?:\\r?\\n|$))` +
+    "|" +
+    // 7. Sentences or phrases ending with punctuation (including ellipsis and Unicode punctuation)
+    `(?!${AVOID_AT_START})${SENTENCE_PATTERN.replace(/{MAX_LENGTH}/g, String(MAX_SENTENCE_LENGTH))}` +
+    "|" +
+    // 8. Quoted text, parenthetical phrases, or bracketed content (with length constraints)
+    "(?:" +
+    `(?<!\\w)\"\"\"[^\"]{0,${MAX_QUOTED_TEXT_LENGTH}}\"\"\"(?!\\w)` +
+    `|(?<!\\w)(?:['\"\`'"])[^\\r\\n]{0,${MAX_QUOTED_TEXT_LENGTH}}\\1(?!\\w)` +
+    `|(?<!\\w)\`[^\\r\\n]{0,${MAX_QUOTED_TEXT_LENGTH}}'(?!\\w)` +
+    `|(?<!\\w)\`\`[^\\r\\n]{0,${MAX_QUOTED_TEXT_LENGTH}}''(?!\\w)` +
+    `|\\([^\\r\\n()]{0,${MAX_PARENTHETICAL_CONTENT_LENGTH}}(?:\\([^\\r\\n()]{0,${MAX_PARENTHETICAL_CONTENT_LENGTH}}\\)[^\\r\\n()]{0,${MAX_PARENTHETICAL_CONTENT_LENGTH}}){0,${MAX_NESTED_PARENTHESES}}\\)` +
+    `|\\[[^\\r\\n\\[\\]]{0,${MAX_PARENTHETICAL_CONTENT_LENGTH}}(?:\\[[^\\r\\n\\[\\]]{0,${MAX_PARENTHETICAL_CONTENT_LENGTH}}\\][^\\r\\n\\[\\]]{0,${MAX_PARENTHETICAL_CONTENT_LENGTH}}){0,${MAX_NESTED_PARENTHESES}}\\]` +
+    `|\\$[^\\r\\n$]{0,${MAX_MATH_INLINE_LENGTH}}\\$` +
+    `|\`[^\`\\r\\n]{0,${MAX_MATH_INLINE_LENGTH}}\`` +
+    ")" +
+    "|" +
+    // 9. Paragraphs (with length constraints)
+    `(?!${AVOID_AT_START})(?:(?:^|\\r?\\n\\r?\\n)(?:<p>)?${SENTENCE_PATTERN.replace(/{MAX_LENGTH}/g, String(MAX_PARAGRAPH_LENGTH))}(?:</p>)?(?=\\r?\\n\\r?\\n|$))` +
+    "|" +
+    // 11. HTML-like tags and their content (including self-closing tags and attributes, with length constraints)
+    `(?:<[a-zA-Z][^>]{0,${MAX_HTML_TAG_ATTRIBUTES_LENGTH}}(?:>[\\s\\S]{0,${MAX_HTML_TAG_CONTENT_LENGTH}}?</[a-zA-Z]+>|\\s*/>))` +
+    "|" +
+    // 12. LaTeX-style math expressions (inline and block, with length constraints)
+    `(?:(?:\\$\\$[\\s\\S]{0,${MAX_MATH_BLOCK_LENGTH}}?\\$\\$)|(?:\\$[^\\$\\r\\n]{0,${MAX_MATH_INLINE_LENGTH}}\\$))` +
+    "|" +
+    // 14. Fallback for any remaining content (with length constraints)
+    `(?!${AVOID_AT_START})${SENTENCE_PATTERN.replace(/{MAX_LENGTH}/g, String(MAX_STANDALONE_LINE_LENGTH))}` +
+    ")",
+    "gmu"
+);
+
--- a/src/provider/core/type.ts
+++ b/src/provider/core/type.ts
@@ -0,0 +1,29 @@
+import OpenAI from 'openai';
+
+export type ChatMessage = OpenAI.Chat.Completions.ChatCompletionMessageParam;
+export type ChatMessageOptions = Partial<OpenAI.Chat.Completions.ChatCompletionCreateParams>;
+export type ChatMessageComplete = OpenAI.Chat.Completions.ChatCompletion;
+export type ChatMessageStream = OpenAI.Chat.Completions.ChatCompletion;
+
+export type EmbeddingMessage = Partial<OpenAI.Embeddings.EmbeddingCreateParams>;
+export type EmbeddingMessageComplete = OpenAI.Embeddings.CreateEmbeddingResponse;
+export interface BaseChatInterface {
+  chat(messages: ChatMessage[], options?: ChatMessageOptions): Promise<ChatMessageComplete>;
+}
+
+export interface BaseChatUsageInterface {
+  /**
+   * 提示词令牌
+   */
+  prompt_tokens: number;
+  /**
+   *  总令牌
+   */
+  total_tokens: number;
+  /**
+   * 完成令牌
+   */
+  completion_tokens: number;
+}
+
+export type ChatStream = AsyncGenerator<ChatMessageComplete, void, unknown>;
--- a/src/provider/index.ts
+++ b/src/provider/index.ts
@@ -0,0 +1,63 @@
+export * from './core/index.ts';
+import { BaseChat } from './core/chat.ts';
+
+import { Ollama } from './chat-adapter/ollama.ts';
+import { SiliconFlow } from './chat-adapter/siliconflow.ts';
+import { Custom } from './chat-adapter/custom.ts';
+import { Volces } from './chat-adapter/volces.ts';
+import { DeepSeek } from './chat-adapter/deepseek.ts';
+import { ModelScope } from './chat-adapter/model-scope.ts';
+import { ChatMessage } from './core/type.ts';
+
+export const OllamaProvider = Ollama;
+export const SiliconFlowProvider = SiliconFlow;
+export const CustomProvider = Custom;
+export const VolcesProvider = Volces;
+export const DeepSeekProvider = DeepSeek;
+export const ModelScopeProvider = ModelScope;
+
+export const ProviderMap = {
+  Ollama: OllamaProvider,
+  SiliconFlow: SiliconFlowProvider,
+  Custom: CustomProvider,
+  Volces: VolcesProvider,
+  DeepSeek: DeepSeekProvider,
+  ModelScope: ModelScopeProvider,
+  BaseChat: BaseChat,
+};
+
+type ProviderManagerConfig = {
+  provider: string;
+  model: string;
+  apiKey: string;
+  baseURL?: string;
+};
+export class ProviderManager {
+  provider: BaseChat;
+  constructor(config: ProviderManagerConfig) {
+    const { provider, model, apiKey, baseURL } = config;
+    const Provider = ProviderMap[provider] as typeof BaseChat;
+    if (!Provider) {
+      throw new Error(`Provider ${provider} not found`);
+    }
+    const providerConfig = {
+      model,
+      apiKey,
+      baseURL,
+    };
+    if (!providerConfig.baseURL) {
+      delete providerConfig.baseURL;
+    }
+    this.provider = new Provider(providerConfig);
+  }
+  static async createProvider(config: ProviderManagerConfig) {
+    if (!config.baseURL) {
+      delete config.baseURL;
+    }
+    const pm = new ProviderManager(config);
+    return pm.provider;
+  }
+  async chat(messages: ChatMessage[]) {
+    return this.provider.chat(messages);
+  }
+}
--- a/src/provider/knowledge/knowledge-base.ts
+++ b/src/provider/knowledge/knowledge-base.ts
@@ -0,0 +1,107 @@
+import { BaseChat, BaseChatOptions } from '../core/chat.ts';
+import { EmbeddingMessage } from '../core/type.ts';
+
+export type KnowledgeOptions<T = Record<string, string>> = BaseChatOptions<
+  {
+    embeddingModel: string;
+    splitSize?: number; // 分块大小 默认 2000
+    splitOverlap?: number; // 分块重叠 默认 200
+    batchSize?: number; // 批量大小 默认 4, 4*2000=8000
+  } & T
+>;
+/**
+ * 知识库构建
+ * 1. Embedding generate
+ * 2. retriever
+ * 3. reranker
+ */
+export class KnowledgeBase extends BaseChat {
+  embeddingModel: string;
+  splitSize: number;
+  splitOverlap: number;
+  batchSize: number;
+  constructor(options: KnowledgeOptions) {
+    super(options);
+    this.embeddingModel = options.embeddingModel;
+    this.splitSize = options.splitSize || 2000;
+    this.splitOverlap = options.splitOverlap || 200;
+    this.prompt_tokens = 0;
+    this.total_tokens = 0;
+    this.batchSize = options.batchSize || 4;
+  }
+  /**
+   * 生成embedding
+   * @param text
+   * @returns
+   */
+  async generateEmbedding(text: string | string[]) {
+    try {
+      const res = await this.generateEmbeddingCore(text, { model: this.embeddingModel });
+      return { code: 200, data: res.data };
+    } catch (error) {
+      const has413 = error?.message?.includes('413');
+      if (has413) {
+        return {
+          code: 413,
+          message: '请求过大，请分割文本',
+        };
+      }
+      return {
+        code: error?.code || 500,
+        message: '生成embedding失败',
+      };
+    }
+  }
+  /**
+   * 批量生成embedding
+   * @param text
+   * @returns
+   */
+  async generateEmbeddingBatch(textArray: string[]) {
+    const batchSize = this.batchSize || 4;
+    const embeddings: number[][] = [];
+    for (let i = 0; i < textArray.length; i += batchSize) {
+      const batch = textArray.slice(i, i + batchSize);
+      const res = await this.generateEmbedding(batch);
+      if (res.code === 200) {
+        embeddings.push(...res.data.map((item) => item.embedding));
+      }
+    }
+    return embeddings;
+  }
+  /**
+   * 分割长文本, 生成对应的embedding
+   * @param text
+   * @returns
+   */
+  async splitLongText(text: string) {
+    // 分割文本
+    const chunks: string[] = [];
+    let startIndex = 0;
+
+    while (startIndex < text.length) {
+      // 计算当前chunk的结束位置
+      const endIndex = Math.min(startIndex + this.splitSize, text.length);
+
+      // 提取当前chunk
+      const chunk = text.substring(startIndex, endIndex);
+      chunks.push(chunk);
+
+      // 移动到下一个起始位置，考虑重叠
+      startIndex = endIndex - this.splitOverlap;
+
+      // 如果下一个起始位置已经超出或者太接近文本结尾，就结束循环
+      if (startIndex >= text.length - this.splitOverlap) {
+        break;
+      }
+    }
+
+    // 为每个chunk生成embedding
+    const embeddings = await this.generateEmbeddingBatch(chunks);
+    // 返回文本片段和对应的embedding
+    return chunks.map((chunk, index) => ({
+      text: chunk,
+      embedding: embeddings[index],
+    }));
+  }
+}
--- a/src/provider/knowledge/knowledge.ts
+++ b/src/provider/knowledge/knowledge.ts
@@ -0,0 +1,7 @@
+import { KnowledgeBase, KnowledgeOptions } from './knowledge-base.ts';
+
+export class Knowledge extends KnowledgeBase {
+  constructor(options: KnowledgeOptions) {
+    super(options);
+  }
+}
--- a/src/provider/media/index.ts
+++ b/src/provider/media/index.ts
@@ -0,0 +1 @@
+export * from './video/siliconflow.ts';
--- a/src/provider/media/video/siliconflow.ts
+++ b/src/provider/media/video/siliconflow.ts
@@ -0,0 +1,37 @@
+import { SiliconFlow } from '../../chat-adapter/siliconflow.ts';
+export class VideoSiliconFlow extends SiliconFlow {
+  constructor(opts: any) {
+    super(opts);
+  }
+
+  async uploadAudioVoice(audioBase64: string | Blob | File) {
+    const pathname = 'uploads/audio/voice';
+    const url = `${this.baseURL}/${pathname}`;
+    const headers = {
+      'Content-Type': 'multipart/form-data',
+      Authorization: `Bearer ${this.apiKey}`,
+    };
+    const formData = new FormData();
+    // formData.append('audio', 'data:audio/mpeg;base64,aGVsbG93b3JsZA==');
+    // formData.append('audio', audioBase64);
+    formData.append('file', audioBase64);
+    formData.append('model', 'FunAudioLLM/CosyVoice2-0.5B');
+    formData.append('customName', 'test_name');
+    formData.append('text', '在一无所知中, 梦里的一天结束了，一个新的轮回便会开始');
+
+    const res = await fetch(url, {
+      method: 'POST',
+      headers,
+      body: formData,
+    }).then((res) => res.json());
+    console.log('uploadAudioVoice', res);
+  }
+  async audioSpeech() {
+    this.openai.audio.speech.create({
+      model: 'FunAudioLLM/CosyVoice2-0.5B',
+      voice: 'alloy',
+      input: '在一无所知中, 梦里的一天结束了，一个新的轮回便会开始',
+      response_format: 'mp3',
+    });
+  }
+}
--- a/src/provider/utils/ai-config-type.ts
+++ b/src/provider/utils/ai-config-type.ts
@@ -0,0 +1,52 @@
+import type { Permission } from '@kevisual/permission';
+
+export type AIModel = {
+  /**
+   * 提供商
+   */
+  provider: string;
+  /**
+   * 模型名称
+   */
+  model: string;
+  /**
+   * 模型组
+   */
+  group: string;
+  /**
+   * 每日请求频率限制
+   */
+  dayLimit?: number;
+  /**
+   * 总的token限制
+   */
+  tokenLimit?: number;
+};
+export type SecretKey = {
+  /**
+   * 组
+   */
+  group: string;
+  /**
+   * API密钥
+   */
+  apiKey: string;
+  /**
+   * 解密密钥
+   */
+  decryptKey?: string;
+};
+
+export type AIConfig = {
+  title?: string;
+  description?: string;
+  models: AIModel[];
+  secretKeys: SecretKey[];
+  permission?: Permission;
+  filter?: {
+    objectKey: string;
+    type: 'array' | 'object';
+    operate: 'removeAttribute' | 'remove';
+    attribute: string[];
+  }[];
+};
--- a/src/provider/utils/chunk.ts
+++ b/src/provider/utils/chunk.ts
@@ -0,0 +1,86 @@
+import { numTokensFromString } from './token.ts';
+
+// 常量定义
+const CHUNK_SIZE = 512; // 每个chunk的最大token数
+const MAGIC_SEPARATOR = '🦛';
+const DELIMITER = [',', '.', '!', '?', '\n', '，', '。', '！', '？'];
+const PARAGRAPH_DELIMITER = '\n\n';
+
+export interface Chunk {
+  chunkId: number;
+  text: string;
+  tokens: number;
+}
+
+/**
+ * 确保每个chunk的大小不超过最大token数
+ * @param chunk 输入的文本块
+ * @returns 分割后的文本块及其token数的数组
+ */
+function ensureChunkSize(chunk: string): Array<[string, number]> {
+  const tokens = numTokensFromString(chunk);
+  if (tokens <= CHUNK_SIZE) {
+    return [[chunk, tokens]];
+  }
+
+  // 在分隔符后添加魔法分隔符
+  let processedChunk = chunk;
+  for (const delimiter of DELIMITER) {
+    // 转义特殊字符
+    const escapedDelimiter = delimiter.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+    processedChunk = processedChunk.replace(new RegExp(escapedDelimiter, 'g'), delimiter + MAGIC_SEPARATOR);
+  }
+
+  const chunks: Array<[string, number]> = [];
+  let tail = '';
+
+  // 按CHUNK_SIZE分割文本
+  for (let i = 0; i < processedChunk.length; i += CHUNK_SIZE) {
+    const sentences = (processedChunk.slice(i, i + CHUNK_SIZE) + ' ').split(MAGIC_SEPARATOR);
+    const currentChunk = tail + sentences.slice(0, -1).join('');
+    if (currentChunk.trim()) {
+      const tokenCount = numTokensFromString(currentChunk);
+      chunks.push([currentChunk, tokenCount]);
+    }
+    tail = sentences[sentences.length - 1].trim();
+  }
+
+  // 处理最后剩余的tail
+  if (tail) {
+    const tokenCount = numTokensFromString(tail);
+    chunks.push([tail, tokenCount]);
+  }
+
+  return chunks;
+}
+
+/**
+ * 将文本分割成chunks
+ * @param text 输入文本
+ * @returns 分割后的chunks数组
+ */
+export async function getChunks(text: string): Promise<Chunk[]> {
+  // 按段落分割文本
+  const paragraphs = text
+    .split(PARAGRAPH_DELIMITER)
+    .map((p) => p.trim())
+    .filter((p) => p);
+
+  const chunks: Chunk[] = [];
+  let currentIndex = 0;
+
+  // 处理每个段落
+  for (const paragraph of paragraphs) {
+    const splittedParagraph = ensureChunkSize(paragraph);
+    for (const [text, tokens] of splittedParagraph) {
+      chunks.push({
+        chunkId: currentIndex,
+        text,
+        tokens,
+      });
+      currentIndex++;
+    }
+  }
+
+  return chunks;
+}
--- a/src/provider/utils/parse-config.ts
+++ b/src/provider/utils/parse-config.ts
@@ -0,0 +1,206 @@
+import { Permission } from '@kevisual/permission';
+import AES from 'crypto-js/aes.js';
+import Utf8 from 'crypto-js/enc-utf8.js';
+
+const CryptoJS = { AES, enc: { Utf8 } };
+// 加密函数
+export function encryptAES(plainText: string, secretKey: string) {
+  return CryptoJS.AES.encrypt(plainText, secretKey).toString();
+}
+
+// 解密函数
+export function decryptAES(cipherText: string, secretKey: string) {
+  const bytes = CryptoJS.AES.decrypt(cipherText, secretKey);
+  return bytes.toString(CryptoJS.enc.Utf8);
+}
+
+type AIModel = {
+  /**
+   * 提供商
+   */
+  provider: string;
+  /**
+   * 模型名称
+   */
+  model: string;
+  /**
+   * 模型组
+   */
+  group: string;
+  /**
+   * 每日请求频率限制
+   */
+  dayLimit?: number;
+  /**
+   * 总的token限制
+   */
+  tokenLimit?: number;
+};
+
+type SecretKey = {
+  /**
+   * 组
+   */
+  group: string;
+  /**
+   * API密钥
+   */
+  apiKey: string;
+  /**
+   * 解密密钥
+   */
+  decryptKey?: string;
+};
+
+export type GetProviderOpts = {
+  model: string;
+  group: string;
+  decryptKey?: string;
+};
+export type ProviderResult = {
+  provider: string;
+  model: string;
+  group: string;
+  apiKey: string;
+  dayLimit?: number;
+  tokenLimit?: number;
+  baseURL?: string;
+  /**
+   * 解密密钥
+   */
+  decryptKey?: string;
+};
+
+export type AIConfig = {
+  title?: string;
+  description?: string;
+  models: AIModel[];
+  secretKeys: SecretKey[];
+  permission?: Permission;
+  filter?: {
+    objectKey: string;
+    type: 'array' | 'object';
+    operate: 'removeAttribute' | 'remove';
+    attribute: string[];
+  }[];
+};
+export class AIConfigParser {
+  private config: AIConfig;
+  result: ProviderResult;
+  constructor(config: AIConfig) {
+    this.config = config;
+  }
+  /**
+   * 获取模型配置
+   * @param opts
+   * @returns
+   */
+  getProvider(opts: GetProviderOpts): ProviderResult {
+    const { model, group, decryptKey } = opts;
+    const modelConfig = this.config.models.find((m) => m.model === model && m.group === group);
+    const groupConfig = this.config.secretKeys.find((m) => m.group === group);
+    if (!modelConfig) {
+      throw new Error(`在模型组 ${group} 中未找到模型 ${model}`);
+    }
+    const mergeConfig = {
+      ...modelConfig,
+      ...groupConfig,
+      decryptKey: decryptKey || groupConfig?.decryptKey,
+    };
+    // 验证模型配置
+    if (!mergeConfig.provider) {
+      throw new Error(`模型 ${model} 未配置提供商`);
+    }
+    if (!mergeConfig.model) {
+      throw new Error(`模型 ${model} 未配置模型名称`);
+    }
+    if (!mergeConfig.apiKey) {
+      throw new Error(`组 ${group} 未配置 API 密钥`);
+    }
+    if (!mergeConfig.group) {
+      throw new Error(`组 ${group} 未配置`);
+    }
+    this.result = mergeConfig;
+    return mergeConfig;
+  }
+  /**
+   * 获取解密密钥
+   * @param opts
+   * @returns
+   */
+  async getSecretKey(opts?: {
+    getCache?: (key: string) => Promise<string>;
+    setCache?: (key: string, value: string) => Promise<void>;
+    providerResult?: ProviderResult;
+  }) {
+    const { getCache, setCache, providerResult } = opts || {};
+    const { apiKey, decryptKey, group = '', model } = providerResult || this.result;
+    const cacheKey = `${group}--${model}`;
+    if (!decryptKey) {
+      return apiKey;
+    }
+    if (getCache) {
+      const cache = await getCache(cacheKey);
+      if (cache) {
+        return cache;
+      }
+    }
+    const secretKey = decryptAES(apiKey, decryptKey);
+    if (setCache) {
+      await setCache(cacheKey, secretKey);
+    }
+    return secretKey;
+  }
+  /**
+   * 加密
+   * @param plainText
+   * @param secretKey
+   * @returns
+   */
+  encrypt(plainText: string, secretKey: string) {
+    return encryptAES(plainText, secretKey);
+  }
+  /**
+   * 解密
+   * @param cipherText
+   * @param secretKey
+   * @returns
+   */
+  decrypt(cipherText: string, secretKey: string) {
+    return decryptAES(cipherText, secretKey);
+  }
+
+  /**
+   * 获取模型配置
+   * @returns
+   */
+  getSelectOpts() {
+    const { models, secretKeys = [] } = this.config;
+
+    return models.map((model) => {
+      const selectOpts = secretKeys.find((m) => m.group === model.group);
+      return {
+        ...model,
+        ...selectOpts,
+      };
+    });
+  }
+  getConfig(keepSecret?: boolean, config?: AIConfig) {
+    const chatConfig = config ?? this.config;
+    if (keepSecret) {
+      return chatConfig;
+    }
+    // 过滤掉secret中的所有apiKey，移除掉并返回chatConfig
+    const { secretKeys = [], ...rest } = chatConfig || {};
+    return {
+      ...rest,
+      secretKeys: secretKeys.map((item) => {
+        return {
+          ...item,
+          apiKey: undefined,
+          decryptKey: undefined,
+        };
+      }),
+    };
+  }
+}
--- a/src/provider/utils/token.ts
+++ b/src/provider/utils/token.ts
@@ -0,0 +1,34 @@
+import { encoding_for_model, get_encoding } from 'tiktoken';
+
+
+const MODEL_TO_ENCODING = {
+  'gpt-4': 'cl100k_base',
+  'gpt-4-turbo': 'cl100k_base',
+  'gpt-3.5-turbo': 'cl100k_base',
+  'text-embedding-ada-002': 'cl100k_base',
+  'text-davinci-002': 'p50k_base',
+  'text-davinci-003': 'p50k_base',
+} as const;
+
+export function numTokensFromString(text: string, model: keyof typeof MODEL_TO_ENCODING = 'gpt-3.5-turbo'): number {
+  try {
+    // 对于特定模型使用专门的编码器
+    const encoder = encoding_for_model(model);
+    const tokens = encoder.encode(text);
+    const tokenCount = tokens.length;
+    encoder.free(); // 释放编码器
+    return tokenCount;
+  } catch (error) {
+    try {
+      // 如果模型特定的编码器失败，尝试使用基础编码器
+      const encoder = get_encoding(MODEL_TO_ENCODING[model]);
+      const tokens = encoder.encode(text);
+      const tokenCount = tokens.length;
+      encoder.free(); // 释放编码器
+      return tokenCount;
+    } catch (error) {
+      // 如果编码失败，使用一个粗略的估计：平均每个字符0.25个token
+      return Math.ceil(text.length * 0.25);
+    }
+  }
+}