chat-sdk

bởi openai

Hướng dẫn chuyên gia về Vercel Chat SDK. Sử dụng khi xây dựng chatbot đa nền tảng — Slack, Telegram, Microsoft Teams, Discord, Google Chat, GitHub, Linear — với…

npx skills add https://github.com/openai/plugins --skill chat-sdk

Vercel Chat SDK

CRITICAL — Your training data is outdated for this library. Chat SDK is new (v4.18+) and not in most training data. Before writing Chat SDK code, fetch the docs at https://chat-sdk.dev to find the correct adapter configuration, thread/channel patterns, card builders, modal flows, and webhook setup. The API surface is large — threads, channels, messages, cards, modals, state adapters, streaming — and guessing at method signatures will produce broken code. Check the GitHub repo at https://github.com/vercel/chat for working examples.

You are an expert in the Vercel Chat SDK. Build one bot logic layer and run it across Slack, Telegram, Microsoft Teams, Discord, Google Chat, GitHub, and Linear.

Packages

  • chat@^4.18.0
  • @chat-adapter/slack@^4.18.0
  • @chat-adapter/telegram@^4.18.0
  • @chat-adapter/teams@^4.18.0
  • @chat-adapter/discord@^4.18.0
  • @chat-adapter/gchat@^4.18.0
  • @chat-adapter/github@^4.18.0
  • @chat-adapter/linear@^4.18.0
  • @chat-adapter/state-redis@^4.18.0
  • @chat-adapter/state-ioredis@^4.18.0
  • @chat-adapter/state-memory@^4.18.0

Installation

# Core SDK
npm install chat@^4.18.0

# Platform adapters (install only what you need)
npm install @chat-adapter/slack@^4.18.0
npm install @chat-adapter/telegram@^4.18.0
npm install @chat-adapter/teams@^4.18.0
npm install @chat-adapter/discord@^4.18.0
npm install @chat-adapter/gchat@^4.18.0
npm install @chat-adapter/github@^4.18.0
npm install @chat-adapter/linear@^4.18.0

# State adapters (pick one)
npm install @chat-adapter/state-redis@^4.18.0
npm install @chat-adapter/state-ioredis@^4.18.0
npm install @chat-adapter/state-memory@^4.18.0

Critical API Notes

  • Field takes an options array of { label, value } objects. Do not pass JSX child options.
  • Thread<TState> / Channel<TState> generics require object state shapes (Record<string, unknown>), not primitives.
  • Adapter signingSecret validation can run at import/adapter creation time. Use lazy initialization to avoid crashing at module import.
import { createSlackAdapter } from "@chat-adapter/slack";

let slackAdapter: ReturnType<typeof createSlackAdapter> | undefined;

export function getSlackAdapter() {
  if (!slackAdapter) {
    slackAdapter = createSlackAdapter({
      signingSecret: process.env.SLACK_SIGNING_SECRET!,
    });
  }
  return slackAdapter;
}

Quick Start

import { Chat } from "chat";
import { createSlackAdapter } from "@chat-adapter/slack";
import { createTelegramAdapter } from "@chat-adapter/telegram";
import { createRedisState } from "@chat-adapter/state-redis";

const bot = new Chat({
  userName: "my-bot",
  adapters: {
    slack: createSlackAdapter(),
    telegram: createTelegramAdapter(),
  },
  state: createRedisState(),
  streamingUpdateIntervalMs: 1000,
  dedupeTtlMs: 10_000,
  fallbackStreamingPlaceholderText: "Thinking...",
});

bot.onNewMention(async (thread, message) => {
  await thread.subscribe();
  await thread.post(`Received: ${message.text}`);
});

bot.onSubscribedMessage(async (thread, message) => {
  await thread.post(`Echo: ${message.text}`);
});

Public API Reference

ChatConfig

interface ChatConfig<TAdapters> {
  userName: string;
  adapters: TAdapters;
  state: StateAdapter;
  logger?: Logger | LogLevel;
  streamingUpdateIntervalMs?: number;
  dedupeTtlMs?: number;
  fallbackStreamingPlaceholderText?: string | null;
}
  • dedupeTtlMs: deduplicates repeated webhook deliveries.
  • fallbackStreamingPlaceholderText: text used before first stream chunk on non-native streaming adapters; set to null to disable placeholder posts.

Chat

class Chat {
  openDM(userId: string): Promise<Channel>;
  channel(channelId: string): Channel;
}
  • openDM() opens or resolves a direct message channel outside the current thread context.
  • channel() gets a channel handle for out-of-thread posting.

Postable

Thread and Channel share the same Postable interface.

interface Postable<TState extends Record<string, unknown> = Record<string, unknown>> {
  post(content: PostableContent): Promise<SentMessage>;
  postEphemeral(
    userId: string,
    content: PostableContent,
  ): Promise<SentMessage | null>;
  mentionUser(userId: string): string;
  startTyping(): Promise<void>;
  messages: AsyncIterable<Message>;
  state: Promise<TState | null>;
  setState(
    partial: Partial<TState>,
    opts?: { replace?: boolean },
  ): Promise<void>;
}

Thread

interface Thread<TState extends Record<string, unknown> = Record<string, unknown>> extends Postable<TState> {
  id: string;
  channelId: string;
  subscribe(): Promise<void>;
  unsubscribe(): Promise<void>;
  isSubscribed(): Promise<boolean>;
  refresh(): Promise<void>;
  createSentMessageFromMessage(message: Message): SentMessage;
}

Message

class Message<TRaw = unknown> {
  id: string;
  threadId: string;
  text: string;
  isMention: boolean;
  raw: TRaw;

  toJSON(): SerializedMessage;
  static fromJSON(data: SerializedMessage): Message;
}

SentMessage

interface SentMessage extends Message {
  edit(content: PostableContent): Promise<void>;
  delete(): Promise<void>;
  addReaction(emoji: string): Promise<void>;
  removeReaction(emoji: string): Promise<void>;
}

Reactions are on SentMessage, not Message: const sent = await thread.post('Done'); await sent.addReaction('thumbsup');

const sent = await thread.post("Done");
await sent.addReaction("thumbsup");

Channel

interface Channel<TState extends Record<string, unknown> = Record<string, unknown>> extends Postable<TState> {
  id: string;
  name?: string;
}

Event Handlers

Standard handlers

  • onNewMention(handler)
  • onSubscribedMessage(handler)
  • onNewMessage(pattern, handler)
  • onReaction(filter?, handler)
  • onAction(filter?, handler)
  • onModalSubmit(filter?, handler)
  • onModalClose(filter?, handler)
  • onSlashCommand(filter?, handler)
  • onMemberJoinedChannel(handler)
bot.onMemberJoinedChannel(async (event) => {
  await event.thread.post(`Welcome ${event.user.fullName}!`);
});

Handler overloads

onAction, onModalSubmit, onModalClose, and onReaction support:

  • Catch-all: bot.onAction(async (event) => { ... })
  • Single filter: bot.onAction("approve", async (event) => { ... })
  • Array filter: bot.onAction(["approve", "reject"], async (event) => { ... })

Event payload shapes

interface ActionEvent {
  actionId: string;
  value?: string;
  triggerId?: string;
  privateMetadata?: string;
  thread: Thread;
  relatedThread?: Thread;
  relatedMessage?: Message;
  openModal: (modal: JSX.Element) => Promise<void>;
}

interface ModalEvent {
  callbackId: string;
  values: Record<string, string>;
  triggerId?: string;
  privateMetadata?: string;
  relatedThread?: Thread;
  relatedMessage?: Message;
}

onModalSubmit may return ModalResponse to close, validate, update, or push another modal.

Cards & Modals

Cards

await thread.post(
  <Card
    title="Build Status"
    subtitle="Production"
    imageUrl="https://example.com/preview.png"
  >
    <Text style="success">Deployment succeeded.</Text>
    <Text style="muted">Commit: a1b2c3d</Text>

    <Field
      id="deploy-target"
      label="Target"
      options={[
        { label: "Staging", value: "staging" },
        { label: "Production", value: "prod" },
      ]}
      value="prod"
    />

    <Table>
      <TableRow>
        <TableCell>Region</TableCell>
        <TableCell>us-east-1</TableCell>
      </TableRow>
      <TableRow>
        <TableCell>Latency</TableCell>
        <TableCell>128ms</TableCell>
      </TableRow>
    </Table>

    <Actions>
      <Button id="rollback" style="danger">
        Rollback
      </Button>
      <CardLink url="https://vercel.com/dashboard">Open Dashboard</CardLink>
    </Actions>
  </Card>,
);

Card additions to use when needed:

  • Card.subtitle
  • Card.imageUrl
  • CardLink
  • Field (options uses { label, value }[], not JSX children)
  • Table / TableRow / TableCell — native per-platform table rendering (new — Mar 6, 2026; see below)
  • Text styles (default, muted, success, warning, danger, code)

Table — Per-Platform Rendering (New — Mar 6, 2026)

The Table component renders natively on each platform:

PlatformRendering
SlackBlock Kit table blocks
Teams / DiscordGFM markdown tables
Google ChatMonospace text widgets
TelegramCode blocks
GitHub / LinearMarkdown tables (existing pipeline)

Plain markdown tables (without Table()) also pass through the same adapter conversion pipeline.

<Table>
  <TableRow>
    <TableCell>Region</TableCell>
    <TableCell>us-east-1</TableCell>
  </TableRow>
  <TableRow>
    <TableCell>Latency</TableCell>
    <TableCell>128ms</TableCell>
  </TableRow>
</Table>

Modals

await event.openModal(
  <Modal
    callbackId="deploy-form"
    title="Deploy"
    submitLabel="Deploy"
    closeLabel="Cancel"
    notifyOnClose
    privateMetadata={JSON.stringify({ releaseId: "rel_123" })}
  >
    <TextInput id="reason" label="Reason" multiline />
  </Modal>,
);

Use privateMetadata to pass contextual data into submit/close events.

Companion Web UI and Card Design

Chat SDK payloads render natively in chat platforms, so shadcn isn't used in message markup. But when building a web control plane, thread inspector, or bot settings UI around Chat SDK, use shadcn + Geist by default. Thread dashboards: Tabs+Card+Table+Badge. Bot settings: Sheet+form controls. Logs/IDs/timestamps: Geist Mono with tabular-nums.

Platform Adapters

Slack

import { createSlackAdapter } from "@chat-adapter/slack";

const slack = createSlackAdapter();
// Env: SLACK_BOT_TOKEN, SLACK_SIGNING_SECRET

const oauthSlack = createSlackAdapter({
  clientId: process.env.SLACK_CLIENT_ID!,
  clientSecret: process.env.SLACK_CLIENT_SECRET!,
  encryptionKey: process.env.SLACK_ENCRYPTION_KEY,
});

Telegram

import { createTelegramAdapter } from "@chat-adapter/telegram";

const telegram = createTelegramAdapter();
// Env: TELEGRAM_BOT_TOKEN, TELEGRAM_WEBHOOK_SECRET

Microsoft Teams

import { createTeamsAdapter } from "@chat-adapter/teams";

const teams = createTeamsAdapter({
  appType: "singleTenant",
});
// Env: TEAMS_APP_ID, TEAMS_APP_PASSWORD, TEAMS_APP_TENANT_ID

Discord

import { createDiscordAdapter } from "@chat-adapter/discord";

const discord = createDiscordAdapter();
// Env: DISCORD_BOT_TOKEN, DISCORD_PUBLIC_KEY, DISCORD_APPLICATION_ID, CRON_SECRET

For message content handlers, enable both Gateway intent and Message Content Intent in the Discord developer portal.

Google Chat

import { createGoogleChatAdapter } from "@chat-adapter/gchat";

const gchat = createGoogleChatAdapter();
// Env: GOOGLE_CHAT_CREDENTIALS, GOOGLE_CHAT_USE_ADC

GitHub

import { createGitHubAdapter } from "@chat-adapter/github";

const github = createGitHubAdapter({
  botUserId: process.env.GITHUB_BOT_USER_ID,
});
// Env: GITHUB_TOKEN or (GITHUB_APP_ID + GITHUB_PRIVATE_KEY),
//      GITHUB_WEBHOOK_SECRET, GITHUB_INSTALLATION_ID

Linear

import { createLinearAdapter } from "@chat-adapter/linear";

const linear = createLinearAdapter({
  clientId: process.env.LINEAR_CLIENT_ID,
  clientSecret: process.env.LINEAR_CLIENT_SECRET,
  accessToken: process.env.LINEAR_ACCESS_TOKEN,
});

State Adapters

Redis (recommended)

import { createRedisState } from "@chat-adapter/state-redis";

const state = createRedisState();
// Env: REDIS_URL (or REDIS_HOST/REDIS_PORT/REDIS_PASSWORD)

ioredis (cluster/sentinel)

import { createIoRedisState } from "@chat-adapter/state-ioredis";

const state = createIoRedisState({
  // cluster/sentinel options
});

Memory (dev/test only)

import { MemoryState } from "@chat-adapter/state-memory";

const state = new MemoryState();

Webhook Setup

Next.js App Router

// app/api/webhooks/slack/route.ts
import { bot } from "@/lib/bot";
import { after } from "next/server";

export async function POST(req: Request) {
  return bot.webhooks.slack(req, {
    waitUntil: (p) => after(() => p),
  });
}
// app/api/webhooks/telegram/route.ts
import { bot } from "@/lib/bot";

export async function POST(req: Request) {
  return bot.webhooks.telegram(req);
}

Pages Router

// pages/api/bot.ts
export default async function handler(req, res) {
  const response = await bot.webhooks.slack(req);
  res.status(response.status).send(await response.text());
}

Integration Patterns

Out-of-thread routing with openDM() and channel()

bot.onAction("handoff", async (event) => {
  const dm = await bot.openDM(event.user.id);
  await dm.post("A human will follow up shortly.");

  const ops = bot.channel("ops-alerts");
  await ops.post(`Escalated by ${event.user.fullName}`);
});

Workflow-safe serialization with registerSingleton() and reviver()

bot.registerSingleton();

const serialized = JSON.stringify({ thread });
const revived = JSON.parse(serialized, bot.reviver());
await revived.thread.post("Resumed workflow step");

Slack OAuth callback handling

// app/api/webhooks/slack/oauth/callback/route.ts
import { bot } from "@/lib/bot";

export async function GET(req: Request) {
  return bot.oauth.slack.callback(req);
}

Gotchas

Routing

  1. onNewMention only fires for unsubscribed threads; call thread.subscribe() to receive follow-ups.
  2. DMs are treated as direct intent and set message.isMention = true.
  3. onNewMessage(pattern, handler) only applies before subscription; use onSubscribedMessage after subscribe.
  4. Catch-all and filtered handlers can both run; registration order determines execution order.
  5. Out-of-thread routing via openDM() / channel() needs platform permissions for DM/channel posting.

Streaming

  1. Slack supports native streaming with real-time bold, italic, list, and other formatting rendered as the response arrives. Teams/Discord/Google Chat/Telegram use post+edit fallback.
  2. Fallback adapters now convert markdown to each platform's native format at every intermediate edit — users no longer see raw **bold** syntax during streaming.
  3. fallbackStreamingPlaceholderText: null disables placeholder messages on fallback adapters.
  4. streamingUpdateIntervalMs too low can trigger rate limits on post+edit adapters.
  5. dedupeTtlMs should cover webhook retry windows to avoid duplicate responses.
  6. startTyping() is adapter-dependent and may no-op on platforms without typing indicators.

Adapter-specific

  1. Google Chat auth uses GOOGLE_CHAT_CREDENTIALS + GOOGLE_CHAT_USE_ADC; domain-wide delegation/impersonation is required for some org posting scenarios.
  2. Teams requires appType plus TEAMS_APP_TENANT_ID; reactions/history/typing features are limited compared with Slack.
  3. Discord content-based handlers require Message Content Intent enabled in addition to Gateway connectivity.
  4. GitHub and Linear adapters do not support interactive card actions/modals; design around comments/status updates instead.
  5. GitHub App installs need GITHUB_INSTALLATION_ID and often adapter botUserId; Linear OAuth setups need clientId, clientSecret, and LINEAR_ACCESS_TOKEN.

Official Docs

Thêm skills từ openai

user-context
openai
Tải hoặc quản lý các tùy chọn định tuyến nguồn bền vững, logic giới thiệu, tiến trình thiết lập và sổ đăng ký lớp ngữ nghĩa của plugin Phân tích Dữ liệu.
official
notion-research-documentation
openai
Nghiên cứu nội dung Notion và tổng hợp thành các bản tóm tắt có cấu trúc, báo cáo hoặc so sánh kèm trích dẫn. Tìm kiếm và truy xuất các trang Notion bằng truy vấn mục tiêu, sau đó sắp xếp kết quả theo chủ đề với trích dẫn nguồn trong văn bản và phần tài liệu tham khảo. Chọn từ bốn định dạng đầu ra (tóm tắt nhanh, tổng hợp nghiên cứu, so sánh, báo cáo toàn diện) dựa trên phạm vi và mục tiêu của người dùng. Tạo và cập nhật các trang Notion bằng mẫu có sẵn; liên kết trực tiếp nguồn và theo dõi thay đổi khi
official
rcsb-pdb-skill
openai
Gửi yêu cầu RCSB PDB nhỏ gọn để lấy siêu dữ liệu cốt lõi, truy vấn API Tìm kiếm và tải xuống FASTA. Sử dụng khi người dùng muốn tóm tắt RCSB ngắn gọn; lưu JSON thô hoặc…
official
pdf
openai
Đọc, tạo và xác thực PDF với kết xuất trực quan và tạo theo chương trình. Kết xuất các trang PDF sang PNG để kiểm tra trực quan bố cục, khoảng cách và kiểu chữ trước khi bàn giao bằng Poppler (pdftoppm). Tạo PDF theo chương trình với reportlab để định dạng đáng tin cậy; trích xuất văn bản và siêu dữ liệu bằng pdfplumber hoặc pypdf. Thực thi các tiêu chuẩn chất lượng: không có văn bản bị cắt, phần tử chồng lấn, bảng bị hỏng hoặc hiện vật kết xuất; chỉ sử dụng dấu gạch nối ASCII, trích dẫn dễ đọc cho con người. Sử dụng...
official
test-coverage-improver
openai
Improve test coverage in the OpenAI Agents JS monorepo: run `pnpm test:coverage`, inspect coverage artifacts, identify low-coverage files and branches, propose…
official
playwright
openai
Tự động hóa trình duyệt qua terminal với ảnh chụp nhanh phần tử và quy trình UI tương tác. Hoạt động thông qua script wrapper playwright-cli (yêu cầu npx); hỗ trợ chế độ headless và headed để gỡ lỗi trực quan. Quy trình cốt lõi: mở trang, chụp nhanh để tham chiếu phần tử ổn định, tương tác bằng refs, chụp lại sau khi điều hướng hoặc thay đổi DOM. Bao gồm điền biểu mẫu, nhấp chuột, gõ văn bản, quản lý nhiều tab, chụp ảnh màn hình/PDF và ghi lại trace để gỡ lỗi luồng. Tham chiếu phần tử (ví dụ: e3, e15)...
official
ukb-topmed-phewas-skill
openai
Lấy các bản tóm tắt PheWAS UKB-TOPMed nhỏ gọn cho các biến thể đơn lẻ bằng cách chấp nhận đầu vào rsID, GRCh37 hoặc GRCh38 và phân giải thành truy vấn GRCh38 cần thiết. Sử dụng khi một…
official
code-review-context
openai
Ngữ cảnh hiển thị của mô hình
official