@delphi-ai/多租户聊天

多租户运行时间管理器 Vercel 聊天SDK。管理每次连接 Chat 懒惰制造、LRU驱逐和 AsyncLocalStorage- 基于连接的上下文.

安装

pnpm add @delphi-ai/multi-tenant-chat

# Peer dependency
pnpm add chat

使用量

import { MultiTenantChat, type ChatConnection } from "@delphi-ai/multi-tenant-chat";

interface MyConnection extends ChatConnection {
  ownerId: string;
}

const manager = new MultiTenantChat<MyConnection>({
  async resolveConnection({ platform, connectionKey, request }) {
    // Look up connection from your database
    return { id: connectionKey, platform, ownerId: "user-123" };
  },

  async resolveCredentials(connection) {
    // Return platform-specific credentials
    return { botToken: "xoxb-..." };
  },

  state: yourStateAdapter,

  platforms: {
    slack: (credentials) => createSlackAdapter(credentials),
    telegram: (credentials) => createTelegramAdapter(credentials),
  },

  async onDirectMessage(ctx, thread, message) {
    await thread.post("Hello from multi-tenant chat!");
  },
});

// In your webhook handler:
export async function POST(request: Request) {
  return manager.handleWebhook(request, {
    platform: "telegram",
    connectionKey: "connection-id",
  });
}

API 密码

MultiTenantChat<T extends ChatConnection>

主类管理每个连接聊天实例.

配置 :

• resolveConnection 解决平台+关键+请求的连接
• resolveCredentials ——返回连接证书
• state a 包括: StateAdapter (在所有连接中共享,密钥为自动前缀)
• platforms ——适配器厂站名地图
• onDirectMessage 编号 : onMention 编号 : onSubscribedMessage ——信息处理器
• onError – 全局错误处理器
• runtimeCacheMax 最大缓存聊天例( 默认 1000)

方法:

• handleWebhook(request, { platform, connectionKey }) —— 向正确的查特实例传递一个网点
• destroyRuntime(connectionId) ——拆掉缓存运行时间

getConnection<T>()

获取当前连接 AsyncLocalStorage 上下文(在处理器内).

createPrefixedState(base, prefix)

环绕任意 StateAdapter 与每个连接键名间隔.

RuntimeCache

LRU 活性缓存 Chat 出现懒惰制造和驱逐的情况.

许可证

麻省理工学院