快速开始

安装

npm install @ai-hivemind/core ai @ai-sdk/openai

Hive-Mind 依赖 Vercel AI SDK，LLM Provider 包是 peer dependency，按需安装。

5 分钟上手

1. 创建技能

创建 skills/translator/SKILL.md：

---
name: translator
description: Translate text between languages
metadata:
  tags: [translate, i18n, multilingual]
---

# Translator

Translate the user's text to the requested target language.
Keep the original formatting and tone.

2. 编写代码

import { createHiveMind } from '@ai-hivemind/core';
import { openai } from '@ai-sdk/openai';

const hive = createHiveMind({
  models: { default: openai('gpt-4o') },
  skills: [{ type: 'local', path: './skills' }],
});

// 自动路由到 translator 技能
const result = await hive.run({
  message: '翻译成英文：今天天气真好',
});

console.log(result.text);
console.log(result.activatedSkills); // ['translator']
console.log(result.usage);          // { promptTokens: 461, ... }

3. 流式输出

const stream = await hive.stream({
  message: '翻译成日文：你好世界',
});

for await (const chunk of stream) {
  process.stdout.write(chunk);
}

环境变量

# .env
OPENAI_API_KEY=sk-proj-xxxxxxxxxxxx

Vercel AI SDK Provider 自动从对应的环境变量中读取 API Key，无需在代码中硬编码。

使用 OpenRouter

如果想通过 OpenRouter 访问多种模型：

npm install @ai-sdk/openai

import { createOpenAI } from '@ai-sdk/openai';

const openrouter = createOpenAI({
  baseURL: 'https://openrouter.ai/api/v1',
  apiKey: process.env.OPENROUTER_API_KEY,
});

const hive = createHiveMind({
  models: {
    default: openrouter('openai/gpt-4o-mini'),
    reasoning: openrouter('anthropic/claude-sonnet-4-20250514'),
  },
  skills: [{ type: 'local', path: './skills' }],
});

下一步

• 核心概念 — 理解三阶段渐进式加载
• 编写技能 — 学习 SKILL.md 格式和 x-hive 扩展
• API 参考 — 完整配置选项