flat: 暂存
This commit is contained in:
Apcallover
329
src/main/ipc/workflow.ts
Normal file
329
src/main/ipc/workflow.ts
Normal file
@@ -0,0 +1,329 @@
|
||||
import { ipcMain, BrowserWindow } from "electron";
|
||||
import { spawn } from "child_process";
|
||||
import { showPrompt } from "../utils/tools";
|
||||
import { preload, indexHtml, ELECTRON_RENDERER_URL } from "../config";
|
||||
import os from "os";
|
||||
import http from "http";
|
||||
let InstallWindows: BrowserWindow | null = null;
|
||||
|
||||
export function setupWorkflowHandlers() {
|
||||
let lastJobSummary = "这是我们今天介绍的第一个岗位";
|
||||
|
||||
// 打开安装窗口
|
||||
ipcMain.handle("open-install-window", async (_, args) => {
|
||||
try {
|
||||
if (InstallWindows) {
|
||||
InstallWindows.focus();
|
||||
showPrompt("下载已打开", "info");
|
||||
return { success: true };
|
||||
}
|
||||
const { width, height, path } = args;
|
||||
let installUrl = `${ELECTRON_RENDERER_URL}/#/${path}`;
|
||||
console.log(installUrl);
|
||||
InstallWindows = new BrowserWindow({
|
||||
title: "模型下载",
|
||||
width,
|
||||
height,
|
||||
minimizable: false, // 是否可以最小化
|
||||
maximizable: false, // 是否可以最小化
|
||||
closable: true, // 窗口是否可关闭
|
||||
alwaysOnTop: false, // 窗口是否永远在别的窗口的上面
|
||||
webPreferences: {
|
||||
preload,
|
||||
nodeIntegration: true,
|
||||
contextIsolation: false,
|
||||
},
|
||||
});
|
||||
// InstallWindows.webContents.openDevTools();
|
||||
InstallWindows.on("closed", () => {
|
||||
InstallWindows = null;
|
||||
});
|
||||
if (ELECTRON_RENDERER_URL) {
|
||||
InstallWindows.loadURL(installUrl);
|
||||
} else {
|
||||
InstallWindows.loadFile(indexHtml, { hash: `/${path}` });
|
||||
}
|
||||
return { success: true };
|
||||
} catch (error: any) {
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
// 监听来自渲染器进程的 'install-ollama' 事件
|
||||
ipcMain.handle("install-ollama-and-model", async (event) => {
|
||||
const webContents = event.sender; // 获取发送事件的窗口
|
||||
const platform = os.platform();
|
||||
const modelToPull = "qwen3:8b";
|
||||
|
||||
const sendStatus = (status) => {
|
||||
if (webContents && !webContents.isDestroyed()) {
|
||||
webContents.send("install-progress", { status });
|
||||
}
|
||||
};
|
||||
|
||||
try {
|
||||
sendStatus("Checking Ollama installation...");
|
||||
try {
|
||||
await streamCommand(
|
||||
"ollama",
|
||||
["-v"],
|
||||
webContents,
|
||||
"install-progress",
|
||||
);
|
||||
sendStatus("Ollama is already installed.");
|
||||
} catch (error) {
|
||||
// Ollama 未安装，执行安装
|
||||
sendStatus("Ollama not found. Starting installation...");
|
||||
|
||||
if (platform === "darwin" || platform === "linux") {
|
||||
// macOS / Linux - 使用官方的 curl 脚本
|
||||
const installCommand =
|
||||
"curl -fsSL https://ollama.com/install.sh | sh";
|
||||
await streamCommand(
|
||||
"sh",
|
||||
["-c", installCommand],
|
||||
webContents,
|
||||
"install-progress",
|
||||
);
|
||||
} else if (platform === "win32") {
|
||||
// Windows - 使用 PowerShell 下载并静默安装
|
||||
const psScript = `
|
||||
$ProgressPreference = 'SilentlyContinue';
|
||||
$tempPath = [System.IO.Path]::Combine($env:TEMP, 'OllamaSetup.exe');
|
||||
Write-Host 'Downloading OllamaSetup.exe...';
|
||||
Invoke-WebRequest -Uri 'https://ollama.com/download/OllamaSetup.exe' -OutFile $tempPath;
|
||||
Write-Host 'Download complete. Starting silent installer...';
|
||||
Start-Process -FilePath $tempPath -ArgumentList '/S' -Wait;
|
||||
Write-Host 'Installation complete. Cleaning up...';
|
||||
Remove-Item $tempPath;
|
||||
Write-Host 'Done.';
|
||||
`;
|
||||
await streamCommand(
|
||||
"powershell",
|
||||
[
|
||||
"-ExecutionPolicy",
|
||||
"Bypass",
|
||||
"-NoProfile",
|
||||
"-Command",
|
||||
psScript,
|
||||
],
|
||||
webContents,
|
||||
"install-progress",
|
||||
);
|
||||
} else {
|
||||
throw new Error(`Unsupported platform: ${platform}`);
|
||||
}
|
||||
sendStatus("Ollama installation complete.");
|
||||
}
|
||||
|
||||
// --- 步骤 2: 拉取模型 ---
|
||||
sendStatus(
|
||||
`Pulling model: ${modelToPull}... (This may take a while)`,
|
||||
);
|
||||
await streamCommand(
|
||||
"ollama",
|
||||
["pull", modelToPull],
|
||||
webContents,
|
||||
"install-progress",
|
||||
);
|
||||
sendStatus(`Model ${modelToPull} pull complete.`);
|
||||
|
||||
return {
|
||||
success: true,
|
||||
message: "Installation and model pull successful.",
|
||||
};
|
||||
} catch (error: any) {
|
||||
console.error(error);
|
||||
sendStatus(`Error: ${error.message}`);
|
||||
return { success: false, message: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
// 将整个工作流封装在 IPC Handler 中
|
||||
ipcMain.handle("run-job-workflow", async (_, userQuery) => {
|
||||
let currentJobData = userQuery || {};
|
||||
|
||||
let answerText = "";
|
||||
|
||||
try {
|
||||
console.log("工作流: 正在调用 Ollama 生成脚本...");
|
||||
|
||||
const systemPromptTemplate = `# 角色 (Role) \n你是一个顶级的招聘KOL和直播带岗专家。你的风格是：专业、中立、风趣，能一针见血地分析岗位优劣。你不是一个AI助手，你就是这个角色。 \n\n# 上下文 (Context) \n我正在运行一个自动化工作流。我会\"一个一个\"地喂给你岗位数据。\n\n # 任务 (Task) \n你的任务是执行以下两个操作，并严格按照“输出格式”返回内容：\n1. 生成口播稿：根据【输入数据A】(上一个岗位的摘要) 和【输入数据B】(当前岗位的JSON)，生成一段完整的、约90秒的口播稿。\n2. 生成新摘要：为【输入数据B】的“当前岗位”生成一个简短的摘要（例如：XX公司的XX岗），以便在下一次调用时使用。\n\n# 核心指令 (Core Instruction)\n### 口播稿的生成规则 (Rules for the Script) \n1. 衔接：口播稿必须以一个自然的“过渡句”开头（基于【输入数据A】）。 特殊情况：如果【输入数据A】是“这是我们今天介绍的第一个岗位。”，则开头应是“热场”或“总起”，而不是衔接。 \n2. 内容：必须介绍岗位名称 \`jobTitle\`。 \n3. 提炼：从 \`jobLocation\`, \`companyName\`, \`education\`,\`experience\`，\`scale\` 中提炼“亮点 (Pro)”。 \n4. 翻译：用“人话”翻译 \`description\`。 \n5. 视角：你是在“评测”这个岗位，而不是在“推销”。\n\n### 口播稿的纯文本要求 (Pure Text Rules for the Script ONLY) \n**[重要]** 以下规则 *仅适用于* “口播稿”部分，不适用于“新摘要”部分： \n1. 绝不包含任何Markdown格式 (\`**\`, \`#\`)。 \n2. 绝不包含任何标签、括号或元数据 (\`[]\`, \`()\`)。 \n3. 绝不包含任何寒暄、问候、或自我介绍 (例如 \"你好\", \"当然\")。 \n4. 必须是可以直接朗读的、完整的、流畅的纯文本。\n\n# 输入数据 (Input Data)\n\n## 输入数据A (上一个岗位摘要) \n${lastJobSummary}\n\n## 输入数据B (当前岗位JSON)\n\`\`\`json\n${JSON.stringify(currentJobData, null, 2)}\n\`\`\`\n\n# 输出格式 (Output Format)\n**[绝对严格的指令]** \n你必须严格按照下面这个“两部分”格式输出，使用 \`---NEXT_SUMMARY---\` 作为唯一的分隔符。 绝不在分隔符之外添加任何多余的文字、解释或Markdown。 \n\n[这里是AI生成的、符合上述所有“纯文本要求”的完整口播稿] \n---NEXT_SUMMARY--- \n[这里是AI为“当前岗位”生成的简短新摘要]`;
|
||||
|
||||
answerText = await runOllamaNonStream(
|
||||
systemPromptTemplate,
|
||||
"qwen3:8b",
|
||||
);
|
||||
|
||||
if (!answerText) {
|
||||
throw new Error("Ollama 返回为空");
|
||||
}
|
||||
} catch (e) {
|
||||
return "抱歉，AI 模型在生成脚本时出错。";
|
||||
}
|
||||
|
||||
try {
|
||||
console.log("工作流: 正在解析 AI 输出...");
|
||||
|
||||
let script = "抱歉，AI没有按预定格式返回脚本，请稍后重试。";
|
||||
let summary = "这是我们今天介绍的第一个岗位。"; // 这是一个安全的“重置”摘要
|
||||
|
||||
if (answerText && typeof answerText === "string") {
|
||||
const parts = answerText.split("---NEXT_SUMMARY---");
|
||||
|
||||
if (parts[0] && parts[0].trim() !== "") {
|
||||
script = parts[0].trim();
|
||||
}
|
||||
|
||||
if (parts[1] && parts[1].trim() !== "") {
|
||||
summary = parts[1].trim();
|
||||
}
|
||||
}
|
||||
|
||||
console.log("工作流: 正在更新状态...");
|
||||
lastJobSummary = summary; // 关键：更新主进程中的状态
|
||||
|
||||
console.log("工作流: 完成，返回口播稿。");
|
||||
return { success: true, data: script }; // 将最终的“口播稿”返回给渲染进程
|
||||
} catch (e: any) {
|
||||
console.error("代码运行或变量更新节点出错:", e);
|
||||
return "抱歉，处理 AI 响应时出错。";
|
||||
}
|
||||
});
|
||||
|
||||
// 检查 Ollama 服务器是否正在运行
|
||||
ipcMain.handle("check-ollama-status", async () => {
|
||||
return await checkOllamaServer();
|
||||
});
|
||||
|
||||
// 处理器：检查服务，如果没运行，就用一个轻量命令唤醒它
|
||||
ipcMain.handle("ensure-ollama-running", async () => {
|
||||
let isRunning = await checkOllamaServer();
|
||||
|
||||
if (isRunning) {
|
||||
return {
|
||||
success: true,
|
||||
message: "Ollama服务器已在运行.",
|
||||
};
|
||||
}
|
||||
|
||||
// 服务未运行。
|
||||
// 我们运行 'ollama ps'。这个命令会与服务通信，
|
||||
// 如果服务没启动，Ollama CLI 会自动启动它。
|
||||
try {
|
||||
await runCommand("ollama", ["ps"]);
|
||||
|
||||
// 给服务一点启动时间 (例如 2 秒)
|
||||
await new Promise((resolve) => setTimeout(resolve, 2000));
|
||||
|
||||
// 再次检查
|
||||
isRunning = await checkOllamaServer();
|
||||
if (isRunning) {
|
||||
return {
|
||||
success: true,
|
||||
message: "Ollama 已在后台启动",
|
||||
};
|
||||
} else {
|
||||
return {
|
||||
success: false,
|
||||
message: "服务启动失败",
|
||||
};
|
||||
}
|
||||
} catch (error: any) {
|
||||
console.error("错误:", error);
|
||||
return {
|
||||
success: false,
|
||||
message: `错误: ${error.message}`,
|
||||
};
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
// 辅助函数：检查 Ollama API 是否可访问
|
||||
function checkOllamaServer() {
|
||||
return new Promise((resolve) => {
|
||||
// 默认端口是 11434
|
||||
const req = http.get("http://127.0.0.1:11434/", (res) => {
|
||||
// "Ollama is running" 的响应码是 200
|
||||
resolve(res.statusCode === 200);
|
||||
});
|
||||
|
||||
// 如果连接被拒绝 (ECONNREFUSED)，则服务未运行
|
||||
req.on("error", () => {
|
||||
resolve(false);
|
||||
});
|
||||
});
|
||||
}
|
||||
|
||||
// 辅助函数：运行一个简单的命令并等待它完成
|
||||
function runCommand(command, args) {
|
||||
return new Promise((resolve, reject) => {
|
||||
const process = spawn(command, args, { shell: true });
|
||||
process.on("close", (code) => {
|
||||
if (code === 0) {
|
||||
resolve(null);
|
||||
} else {
|
||||
reject(new Error(`Command failed with code ${code}`));
|
||||
}
|
||||
});
|
||||
process.on("error", (err) => reject(err));
|
||||
});
|
||||
}
|
||||
// 这是一个非流式的 Ollama 助手函数
|
||||
async function runOllamaNonStream(prompt, model = "qwen3:8b") {
|
||||
try {
|
||||
const response = await fetch("http://127.0.0.1:11434/api/chat", {
|
||||
method: "POST",
|
||||
headers: { "Content-Type": "application/json" },
|
||||
body: JSON.stringify({
|
||||
model: model,
|
||||
messages: [{ role: "user", content: prompt }],
|
||||
stream: false, // 关键：关闭流式
|
||||
}),
|
||||
});
|
||||
|
||||
if (!response.ok) {
|
||||
throw new Error(`Ollama API error: ${response.statusText}`);
|
||||
}
|
||||
|
||||
const data = await response.json();
|
||||
|
||||
// data.message.content 包含了完整的 AI 回复
|
||||
return data.message.content;
|
||||
} catch (error) {
|
||||
console.error("Ollama Chat Error:", error);
|
||||
return null; // 返回 null 以便后续逻辑处理
|
||||
}
|
||||
}
|
||||
|
||||
function streamCommand(command, args, webContents, eventName) {
|
||||
return new Promise((resolve, reject) => {
|
||||
const process = spawn(command, args, { shell: true });
|
||||
|
||||
const send = (channel, data) => {
|
||||
if (webContents && !webContents.isDestroyed()) {
|
||||
webContents.send(channel, data);
|
||||
}
|
||||
};
|
||||
|
||||
process.stdout.on("data", (data) => {
|
||||
send(eventName, { type: "stdout", data: data.toString() });
|
||||
});
|
||||
|
||||
process.stderr.on("data", (data) => {
|
||||
send(eventName, { type: "stderr", data: data.toString() });
|
||||
});
|
||||
|
||||
process.on("close", (code) => {
|
||||
if (code === 0) {
|
||||
resolve(null);
|
||||
} else {
|
||||
reject(new Error(`Process exited with code ${code}`));
|
||||
}
|
||||
});
|
||||
|
||||
process.on("error", (err) => {
|
||||
reject(err);
|
||||
});
|
||||
});
|
||||
}
|
||||
Reference in New Issue
Block a user