mirror of
https://github.com/fofolee/uTools-quickcommand.git
synced 2025-06-06 13:04:10 +08:00
507 lines
13 KiB
JavaScript
507 lines
13 KiB
JavaScript
const DOMPurify = require("dompurify");
|
||
const marked = require("marked");
|
||
|
||
window.aiResponseParser = (content) => {
|
||
const markedContent = marked.parse(content.trim());
|
||
const processedContent = markedContent
|
||
.replace("<p><think>", "<think><p>")
|
||
.replace("</think></p>", "</p></think>")
|
||
.replace("<think>\n\n</think>", "");
|
||
const purifiedContent = DOMPurify.sanitize(processedContent, {
|
||
ADD_TAGS: ["think"],
|
||
});
|
||
return purifiedContent;
|
||
};
|
||
|
||
// 支持的模型类型
|
||
const API_TYPES = {
|
||
OPENAI: "openai",
|
||
OLLAMA: "ollama",
|
||
UTOOLS: "utools",
|
||
};
|
||
|
||
// 角色提示词
|
||
const ROLE_PROMPTS = {
|
||
// 翻译
|
||
translate: `你是一名翻译专家，请将我给你的内容进行翻译，要求：
|
||
1. 无论给的内容长短，请直接翻译，不要进行任何解释
|
||
2. 提供中文时，翻译成地道的英文，符合英文的表达习惯
|
||
3. 提供英文时，翻译成地道的中文，符合中文的表达习惯
|
||
4. 保持原文的专业性和准确性
|
||
5. 保持原文的段落格式
|
||
6. 不要输出原文
|
||
7. 不要使用markdown格式，请直接输出翻译后的内容
|
||
`,
|
||
|
||
// 生成SHELL命令
|
||
shell: `你是一名shell命令专家，请根据我的描述生成 shell 命令，要求：
|
||
1. 命令应当简洁高效
|
||
2. 优先使用常见的命令行工具
|
||
3. 确保命令的安全性和可靠性
|
||
4. 对于复杂操作，添加注释说明
|
||
5. 如果需要多个命令，使用 && 连接或使用脚本格式
|
||
6. 直接输出命令，不要输出任何解释，不要使用markdown格式
|
||
`,
|
||
|
||
// 总结
|
||
summarize: `你是一名总结专家，请总结我给你的内容的要点，要求：
|
||
1. 提取最重要和最有价值的信息
|
||
2. 使用简洁的语言
|
||
3. 按重要性排序
|
||
4. 保持逻辑性和连贯性
|
||
5. 请将所有内容放在一个段落内
|
||
6. 不要输出原文
|
||
7. 不要使用markdown格式
|
||
`,
|
||
|
||
// 润色
|
||
polish: `你是一名文字润色专家，请对我给你的内容进行润色，要求：
|
||
1. 保持原文的核心意思不变
|
||
2. 使用更优美、专业的表达方式
|
||
3. 改善语言流畅度和可读性
|
||
4. 纠正语法错误和不恰当的表达
|
||
5. 保持原文的语言风格（中文/英文）
|
||
6. 保持原文的段落格式
|
||
7. 直接输出润色后的内容，不要解释修改原因
|
||
8. 不要使用markdown格式
|
||
`,
|
||
|
||
// 扩写
|
||
expand: `你是一名文字扩写专家，请对我给你的内容进行扩写，要求：
|
||
1. 在保持原意的基础上扩充内容
|
||
2. 添加相关的细节和例子
|
||
3. 使用生动的描述和丰富的表达
|
||
4. 确保扩写内容逻辑连贯
|
||
5. 适度扩写，避免过度冗长
|
||
6. 保持语言风格的一致性
|
||
7. 直接输出扩写后的内容，不要解释扩写原因
|
||
8. 不要使用markdown格式
|
||
`,
|
||
};
|
||
|
||
// API URL 处理
|
||
const API_ENDPOINTS = {
|
||
[API_TYPES.OPENAI]: {
|
||
chat: "/v1/chat/completions",
|
||
models: "/v1/models",
|
||
},
|
||
[API_TYPES.OLLAMA]: {
|
||
chat: "/api/chat",
|
||
models: "/api/tags",
|
||
},
|
||
};
|
||
|
||
// 构建API URL
|
||
function buildApiUrl(baseUrl, endpoint) {
|
||
if (!baseUrl.endsWith(endpoint)) {
|
||
return baseUrl.replace(/\/?$/, endpoint);
|
||
}
|
||
return baseUrl;
|
||
}
|
||
|
||
// 构建请求配置
|
||
function buildRequestConfig(apiConfig) {
|
||
const config = {
|
||
headers: {
|
||
"Content-Type": "application/json",
|
||
},
|
||
};
|
||
|
||
if (apiConfig.apiType === API_TYPES.OPENAI && apiConfig.apiToken) {
|
||
config.headers["Authorization"] = `Bearer ${apiConfig.apiToken}`;
|
||
}
|
||
|
||
return config;
|
||
}
|
||
|
||
// 构建请求数据
|
||
function buildRequestData(content, apiConfig) {
|
||
const { model } = apiConfig;
|
||
const { prompt, role, context = [] } = content;
|
||
const rolePrompt = ROLE_PROMPTS[role] || role;
|
||
|
||
const roleMessage = rolePrompt
|
||
? [
|
||
{
|
||
role: "system",
|
||
content: rolePrompt,
|
||
},
|
||
]
|
||
: [];
|
||
|
||
// 统一的消息格式处理
|
||
const messages = [
|
||
// 添加系统角色消息（如果有）
|
||
...roleMessage,
|
||
// 添加上下文消息
|
||
...context.map((msg) => ({
|
||
role: msg.role || "user",
|
||
content: msg.content,
|
||
})),
|
||
// 添加当前用户消息
|
||
{
|
||
role: "user",
|
||
content: prompt,
|
||
},
|
||
];
|
||
|
||
return {
|
||
model,
|
||
messages,
|
||
stream: true,
|
||
};
|
||
}
|
||
|
||
// 处理模型列表响应
|
||
function parseModelsResponse(response, apiType) {
|
||
if (apiType === API_TYPES.OPENAI) {
|
||
if (!response.data.data) {
|
||
throw new Error("OpenAI 响应格式错误");
|
||
}
|
||
return response.data.data.map((model) => model.id);
|
||
} else {
|
||
if (!response.data.models) {
|
||
throw new Error("Ollama 响应格式错误");
|
||
}
|
||
return response.data.models.map((model) => model.name);
|
||
}
|
||
}
|
||
|
||
let reasoning_content_start = false;
|
||
function processContentWithReason(response, onStream) {
|
||
if (response.reasoning_content) {
|
||
if (!reasoning_content_start) {
|
||
reasoning_content_start = true;
|
||
onStream("<think>", false);
|
||
}
|
||
onStream(response.reasoning_content, false);
|
||
}
|
||
if (response.content) {
|
||
if (reasoning_content_start) {
|
||
reasoning_content_start = false;
|
||
onStream("</think>", false);
|
||
}
|
||
onStream(response.content, false);
|
||
}
|
||
}
|
||
|
||
// 处理 OpenAI 流式响应
|
||
async function handleOpenAIStreamResponse(line, onStream) {
|
||
if (line.startsWith("data:")) {
|
||
const jsonStr = line.replace(/^data:[ ]*/, "");
|
||
if (jsonStr === "[DONE]") {
|
||
onStream("", true);
|
||
return;
|
||
}
|
||
const json = JSON.parse(jsonStr);
|
||
const response = json.choices[0]?.delta;
|
||
if (response) {
|
||
processContentWithReason(response, onStream);
|
||
}
|
||
}
|
||
}
|
||
|
||
// 处理 Ollama 流式响应
|
||
async function handleOllamaStreamResponse(line, onStream) {
|
||
const json = JSON.parse(line);
|
||
if (json.done) {
|
||
onStream("", true);
|
||
return;
|
||
}
|
||
const response = json.message;
|
||
if (response) {
|
||
processContentWithReason(response, onStream);
|
||
}
|
||
}
|
||
|
||
// 处理 uTools AI 流式响应
|
||
async function handleUToolsAIStreamResponse(response, onStream) {
|
||
processContentWithReason(response, onStream);
|
||
}
|
||
|
||
// 处理流式响应
|
||
async function handleStreamResponse(response, apiConfig, onStream) {
|
||
// 处理 uTools AI 响应
|
||
if (apiConfig.apiType === API_TYPES.UTOOLS) {
|
||
try {
|
||
await handleUToolsAIStreamResponse(response, onStream);
|
||
return { success: true };
|
||
} catch (error) {
|
||
throw error;
|
||
}
|
||
}
|
||
|
||
// 处理其他 API 的流式响应
|
||
const reader = response.body.getReader();
|
||
const decoder = new TextDecoder();
|
||
let buffer = "";
|
||
|
||
try {
|
||
while (true) {
|
||
const { value, done } = await reader.read();
|
||
if (done) break;
|
||
|
||
buffer += decoder.decode(value, { stream: true });
|
||
const lines = buffer.split("\n");
|
||
buffer = lines.pop() || "";
|
||
|
||
for (const line of lines) {
|
||
if (line.trim()) {
|
||
try {
|
||
if (apiConfig.apiType === API_TYPES.OPENAI) {
|
||
await handleOpenAIStreamResponse(line, onStream);
|
||
} else {
|
||
await handleOllamaStreamResponse(line, onStream);
|
||
}
|
||
} catch (e) {
|
||
console.error("解析响应失败:", e);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
// 处理剩余的缓冲区
|
||
if (buffer.trim()) {
|
||
try {
|
||
if (apiConfig.apiType === API_TYPES.OPENAI) {
|
||
await handleOpenAIStreamResponse(buffer, onStream);
|
||
} else {
|
||
await handleOllamaStreamResponse(buffer, onStream);
|
||
}
|
||
} catch (e) {
|
||
console.error("解析剩余响应失败:", e);
|
||
}
|
||
}
|
||
} catch (error) {
|
||
if (error.name === "AbortError") {
|
||
return {
|
||
success: false,
|
||
error: "请求已取消",
|
||
cancelled: true,
|
||
};
|
||
}
|
||
throw error;
|
||
} finally {
|
||
reader.releaseLock();
|
||
}
|
||
|
||
return { success: true };
|
||
}
|
||
|
||
/**
|
||
* AI对话功能
|
||
* @param {Object} content - 对话内容参数
|
||
* @param {Object} apiConfig - API配置参数
|
||
* @param {Object} options - 其他选项
|
||
* @returns {Promise<Object>} 对话响应
|
||
*/
|
||
async function chat(content, apiConfig, options = {}) {
|
||
try {
|
||
const {
|
||
showProcessBar = true,
|
||
onStream = () => {},
|
||
onFetch = () => {},
|
||
} = options;
|
||
|
||
// 验证必要参数
|
||
if (apiConfig.apiType === API_TYPES.UTOOLS) {
|
||
if (!content.prompt || !apiConfig.model) {
|
||
throw new Error("模型名称和提示词不能为空");
|
||
}
|
||
} else {
|
||
if (!apiConfig.apiUrl) {
|
||
throw new Error("API地址不能为空");
|
||
}
|
||
if (!apiConfig.apiUrl || !content.prompt || !apiConfig.model) {
|
||
throw new Error("API地址、模型名称和提示词不能为空");
|
||
}
|
||
}
|
||
|
||
let controller;
|
||
|
||
// 显示进度条
|
||
const processBar = showProcessBar
|
||
? await quickcommand.showProcessBar({
|
||
text: "AI思考中...",
|
||
onClose: () => {
|
||
if (typeof controller !== "undefined") {
|
||
controller.abort();
|
||
}
|
||
},
|
||
})
|
||
: null;
|
||
|
||
// 用于收集完整响应
|
||
let fullResponse = "";
|
||
|
||
// 包装 onStream 回调以收集完整响应并更新进度条
|
||
const streamHandler = (chunk, isDone) => {
|
||
if (!isDone) {
|
||
fullResponse += chunk;
|
||
// 更新进度条显示最新的响应内容
|
||
if (processBar) {
|
||
quickcommand.updateProcessBar(
|
||
{
|
||
text: window.aiResponseParser(fullResponse),
|
||
},
|
||
processBar
|
||
);
|
||
}
|
||
}
|
||
onStream(chunk, isDone);
|
||
};
|
||
|
||
// 处理 uTools AI 请求
|
||
if (apiConfig.apiType === API_TYPES.UTOOLS) {
|
||
try {
|
||
const messages = buildRequestData(content, apiConfig).messages;
|
||
controller = utools.ai(
|
||
{
|
||
model: apiConfig.model,
|
||
messages: messages,
|
||
},
|
||
(chunk) => {
|
||
handleUToolsAIStreamResponse(chunk, streamHandler);
|
||
}
|
||
);
|
||
onFetch(controller);
|
||
|
||
await controller;
|
||
|
||
// 在流式响应完全结束后，发送一个空字符串表示结束
|
||
streamHandler("", true);
|
||
|
||
// 完成时更新进度条并关闭
|
||
if (processBar) {
|
||
quickcommand.updateProcessBar(
|
||
{
|
||
text: "AI响应完成",
|
||
complete: true,
|
||
},
|
||
processBar
|
||
);
|
||
}
|
||
|
||
return {
|
||
success: true,
|
||
result: fullResponse,
|
||
};
|
||
} catch (error) {
|
||
if (error.name === "AbortError") {
|
||
return {
|
||
success: false,
|
||
error: "请求已取消",
|
||
cancelled: true,
|
||
};
|
||
}
|
||
throw error;
|
||
}
|
||
}
|
||
|
||
// 统一使用 fetch 处理其他 API 请求
|
||
controller = new AbortController();
|
||
|
||
onFetch(controller);
|
||
|
||
const url = buildApiUrl(
|
||
apiConfig.apiUrl,
|
||
API_ENDPOINTS[apiConfig.apiType].chat
|
||
);
|
||
const config = buildRequestConfig(apiConfig);
|
||
const requestData = buildRequestData(content, apiConfig);
|
||
|
||
const response = await fetch(url, {
|
||
method: "POST",
|
||
headers: config.headers,
|
||
body: JSON.stringify(requestData),
|
||
signal: controller.signal,
|
||
});
|
||
|
||
if (!response.ok) {
|
||
processBar?.close();
|
||
throw new Error(`HTTP error! status: ${response.status}`);
|
||
}
|
||
|
||
const result = await handleStreamResponse(
|
||
response,
|
||
apiConfig,
|
||
streamHandler
|
||
);
|
||
|
||
// 如果请求被取消，返回取消状态
|
||
if (!result.success) {
|
||
processBar?.close();
|
||
return result;
|
||
}
|
||
|
||
// 完成时更新进度条并关闭
|
||
if (processBar) {
|
||
quickcommand.updateProcessBar(
|
||
{
|
||
text: "AI响应完成",
|
||
complete: true,
|
||
},
|
||
processBar
|
||
);
|
||
}
|
||
|
||
// 返回完整的响应内容
|
||
return {
|
||
success: true,
|
||
result: fullResponse,
|
||
};
|
||
} catch (error) {
|
||
if (error.name === "AbortError") {
|
||
return {
|
||
success: false,
|
||
error: "请求已取消",
|
||
cancelled: true,
|
||
};
|
||
}
|
||
return {
|
||
success: false,
|
||
error: error.response?.data?.error?.message || error.message,
|
||
};
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 获取API支持的模型列表
|
||
* @param {Object} apiConfig - API配置参数
|
||
* @returns {Promise<Object>} 模型列表响应
|
||
*/
|
||
async function getModels(apiConfig) {
|
||
try {
|
||
if (!apiConfig.apiUrl) {
|
||
throw new Error("API地址不能为空");
|
||
}
|
||
|
||
const url = buildApiUrl(
|
||
apiConfig.apiUrl,
|
||
API_ENDPOINTS[apiConfig.apiType].models
|
||
);
|
||
const config = buildRequestConfig(apiConfig);
|
||
|
||
const response = await fetch(url, {
|
||
method: "GET",
|
||
headers: config.headers,
|
||
});
|
||
|
||
if (!response.ok) {
|
||
throw new Error(`HTTP error! status: ${response.status}`);
|
||
}
|
||
|
||
const responseData = await response.json();
|
||
return {
|
||
success: true,
|
||
result: parseModelsResponse({ data: responseData }, apiConfig.apiType),
|
||
};
|
||
} catch (error) {
|
||
return {
|
||
success: false,
|
||
error: error.response?.data?.error?.message || error.message,
|
||
};
|
||
}
|
||
}
|
||
module.exports = { chat, getModels };
|