vercel__chat/packages/adapter-x/README.md

@chat-adapter/x

npm package: @chat-adapter/x

X (Twitter) adapter for Chat SDK, using the X API v2 and the X Activity API. Reply to public mentions, hold DM conversations, post from the bot account, and like posts.

Documentation: chat-sdk.dev/adapters/official/x · Guides: vercel.com/kb/chat-sdk

Installation

pnpm add @chat-adapter/x

Scaffold with the CLI

To scaffold a new X bot with this adapter preselected:

npx create-chat-sdk@latest my-bot --adapter x memory

Visit the adapters directory to see other available official and vendor-official adapters.

Usage

import { Chat } from "chat";
import { createXAdapter } from "@chat-adapter/x";

const bot = new Chat({
  userName: "mybot",
  adapters: {
    x: createXAdapter(),
  },
});

bot.onNewMention(async (thread, message) => {
  await thread.post(`Hi @${message.author.userName}!`);
});

bot.onDirectMessage(async (thread, message) => {
  await thread.post("Hello from X!");
});

When using createXAdapter() without arguments, credentials are auto-detected from environment variables.

X setup

1. Create an X app

1. Go to the X developer portal and create a Project and App
2. Under Keys and tokens, copy the API Key Secret (consumer secret): this becomes X_CONSUMER_SECRET
3. Enable OAuth 2.0 user authentication with the scopes tweet.read, tweet.write, users.read, dm.read, dm.write, like.write, and offline.access
4. Complete the OAuth 2.0 flow for the bot account. Either store the access token as X_USER_ACCESS_TOKEN, or store X_CLIENT_ID plus X_REFRESH_TOKEN to let the adapter manage token refresh

2. Register a webhook

X delivers events through the X Activity API. Set this up once in the X developer console, which handles the auth for you:

1. Register your webhook URL (https://your-domain.com/api/webhooks/x). It must be public HTTPS without a port. X immediately sends a CRC challenge, which the adapter answers automatically
2. Create subscriptions for the events the adapter consumes: post.mention.create, dm.received, and dm.sent (private events, so the bot user must have authorized your app first)

Subscription and webhook management is one-time setup, not adapter runtime. If you script it instead of using the console, the Activity API endpoints are auth-picky and operation-specific and do not fully match the published spec (creating a private-event subscription needed OAuth 1.0a user context in testing, while list and delete used the app-only bearer token), so the console is the simpler path.

3. Environment variables

X_CONSUMER_SECRET=...      # App API key secret, used for webhook CRC and signature verification

# Auth option A: static access token
X_USER_ACCESS_TOKEN=...    # OAuth 2.0 user-context access token for outbound calls

# Auth option B: managed OAuth refresh (recommended for long-running bots)
X_CLIENT_ID=...            # OAuth 2.0 client ID
X_REFRESH_TOKEN=...        # OAuth 2.0 refresh token (requires the offline.access scope)
X_CLIENT_SECRET=...        # Optional, only for confidential clients
X_ENCRYPTION_KEY=...       # Optional, base64 32-byte key to encrypt persisted tokens

X_USER_ID=...              # Bot account user ID. Optional if omitted it is fetched from /2/users/me; the adapter requires a resolvable bot id and fails init otherwise
X_USERNAME=...             # Optional, bot @handle for mention detection (fetched when omitted)
X_API_BASE_URL=...         # Optional, override the X API base URL

Token refresh

X OAuth 2.0 user tokens are short-lived (about two hours). With X_CLIENT_ID and X_REFRESH_TOKEN set, the adapter refreshes the access token before expiry and persists the rotated refresh token in your state adapter, so the bot survives restarts. Set X_ENCRYPTION_KEY to store those tokens AES-256-GCM encrypted.

Alternatively, pass a token provider and plug in your own refresh logic:

import { createXAdapter } from "@chat-adapter/x";

const adapter = createXAdapter({
  userAccessToken: async () => refreshTokenFromMyStore(),
});

Webhook setup

X uses two webhook mechanisms, both handled by the adapter:

1. CRC challenge (GET): X sends a crc_token that the adapter answers with an HMAC-SHA256 response keyed by your consumer secret. X re-validates hourly
2. Event delivery (POST): activity events signed via the x-twitter-webhooks-signature header, verified against the raw request body

// Next.js App Router example
import { bot } from "@/lib/bot";

export async function GET(request: Request) {
  return bot.webhooks.x(request);
}

export async function POST(request: Request) {
  return bot.webhooks.x(request);
}

Features

Messaging

Feature	Supported
Post message	Yes (mention replies and DMs)
Top-level posts	Yes (channel.post on x:public)
Edit message	Posts only (X edit eligibility rules apply)
Delete message	Posts and own DM events
Streaming	Buffered (accumulates then posts once)
Typing indicator	No

Rich content

Feature	Supported
Card format	Plain text fallback
Buttons	No (link buttons render as text)
Tables	ASCII
Modals	No
Image uploads	Yes (png, jpeg, webp; up to 4 per post; also DMs)

Attach images by passing files (or attachments) on the message; the adapter uploads each through X's chunked media endpoints (initialize then append then finalize) and attaches the returned media_ids to the post or DM. A post can carry media with or without text.

await thread.post({
  markdown: "France lead the title race",
  files: [{ data: pngBuffer, filename: "odds.png", mimeType: "image/png" }],
});

Media upload requires the media.write scope on your OAuth 2.0 token, in addition to tweet.write. Mint the token with media.write included or uploads fail with a 403.

Conversations

Feature	Supported
Mentions	Yes (post.mention.create)
DMs	Yes (dm.received / dm.sent)
Reactions	Likes only (emoji.heart or "like")
User lookup	Yes

Message history

Feature	Supported
Fetch messages	DMs via API, posts from cache
Fetch single message	Posts via API, DMs from cache

Thread ID format

x:post:{conversationId}   # public post threads (channel: x:public)
x:dm:{participantUserId}  # direct message with a single user

Examples: x:post:1943467279943467279, x:dm:783214. X DM webhooks carry no conversation id, only participant ids, so DMs are threaded by the other participant's user id: openDM("783214") returns x:dm:783214, and sends route to POST /2/dm_conversations/with/783214/messages. Top-level posts go through channel.post on the x:public channel.

Automation policy

X enforces automation rules. Before deploying a bot:

• get explicit consent before sending automated replies or DMs, and honor opt-outs immediately
• disclose the bot identity in the account profile
• never send bulk, duplicate, or aggressive automated content
• use only the official API (no scraping or browser automation)

The adapter is strict by default: it never streams by post-and-edit, rejects unsupported interactions loudly, and only replies where your handlers decide to.

XChat (encrypted messaging)

XChat is X's encrypted messaging product. It ships from the @chat-adapter/x/chat subpath of this package, so it installs with @chat-adapter/x and adds no dependencies unless you import it.

import { createXchatAdapter } from "@chat-adapter/x/chat";

The XChat crypto stack (@xdevplatform/chat-xdk, @xdevplatform/xdk, juicebox-sdk) is declared as optional peer dependencies. Install them alongside @chat-adapter/x when you use this subpath.

Usage

import { Chat } from "chat";
import { createXchatAdapter } from "@chat-adapter/x/chat";
import { createMemoryState } from "@chat-adapter/state-memory";

const bot = new Chat({
  userName: "mybot",
  adapters: {
    xchat: createXchatAdapter(),
  },
  state: createMemoryState(),
});

// DMs always
bot.onDirectMessage(async (thread, message) => {
  await thread.post(`You said: ${message.text}`);
});

// Group chats when the bot is @mentioned
bot.onNewMention(async (thread, message) => {
  await thread.post(`You said: ${message.text}`);
});

Wire the webhook route (e.g. app/api/webhooks/xchat/route.ts in Next.js):

import { bot } from "@/lib/bot";

export async function GET(request: Request) {
  return bot.webhooks.xchat(request);
}

export async function POST(request: Request) {
  return bot.webhooks.xchat(request);
}

When using createXchatAdapter() without arguments, credentials are auto-detected from environment variables. The bot's @handle is resolved from GET /2/users/me at startup, so mention detection works without any hardcoding.

X Chat setup

1. Register encryption keys

The bot account needs registered public keys and a Juicebox-backed private-key store before it can participate in encrypted conversations:

1. Generate keypairs with chat-xdk (generateKeypairs())
2. Register them via POST /2/users/{id}/public_keys
3. Store the private keys in Juicebox with a PIN (chat.setup(pin))

The adapter unlocks the keys at startup with the same PIN (XCHAT_PIN). Registration is a one-time step per account — see the chat-xdk documentation for the full flow.

2. Create a webhook

Register a webhook URL so the X Activity API can deliver events (docs):

curl -X POST "https://api.x.com/2/webhooks" \
  -H "Authorization: Bearer $APP_BEARER_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://your-domain.com/api/webhooks/xchat"}'

X validates the URL with a CRC challenge (see Webhook route below), so the endpoint must be live before you create the webhook.

Webhooks and activity subscriptions can also be created and managed in the X Developer Portal instead of via the API.

3. Subscribe to chat events

Create activity subscriptions for the bot user with the bot's OAuth2 user token:

curl -X POST "https://api.x.com/2/activity/subscriptions" \
  -H "Authorization: Bearer $BOT_USER_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "event_type": "chat.received",
    "filter": {"user_id": "YOUR_BOT_USER_ID"},
    "tag": "bot-chat-received",
    "webhook_id": "YOUR_WEBHOOK_ID"
  }'

Subscribe to chat.conversation_join as well if you want the bot to post a welcome message when it is added to a group.

Webhook route

X sends two kinds of requests:

1. CRC challenge (GET) — ?crc_token=... must be answered with an HMAC-SHA256 of the token, keyed by your app's consumer secret. Handle this at the route level; it needs no adapter state.
2. Event delivery (POST) — chat events, verified by the adapter via the x-twitter-webhooks-signature header.

import { createHmac } from "node:crypto";
import { bot } from "@/lib/bot";

export async function GET(request: Request) {
  const url = new URL(request.url);
  const crcToken = url.searchParams.get("crc_token");
  if (!crcToken) {
    return new Response("Missing crc_token", { status: 400 });
  }
  const hash = createHmac("sha256", process.env.X_CONSUMER_SECRET!)
    .update(crcToken)
    .digest("base64");
  return Response.json({ response_token: `sha256=${hash}` });
}

export async function POST(request: Request) {
  return bot.webhooks.xchat(request);
}

Configuration

All options are auto-detected from environment variables when not provided.

Option	Required	Description
botToken / accessToken	No*	OAuth2 user access token for the bot account. Auto-detected from XCHAT_BOT_TOKEN or X_ACCESS_TOKEN
pin	No	Juicebox PIN; when set, keys unlock automatically during initialize(). Auto-detected from XCHAT_PIN
consumerSecret	For webhooks	App consumer secret for webhook signature verification. Auto-detected from X_CONSUMER_SECRET. Required to receive webhooks: X signs every POST, so without it every POST is rejected with 401. Polling deployments do not need it
disableWebhookVerification	No	Accept webhook POSTs without verifying their HMAC. Auto-detected from X_DISABLE_WEBHOOK_VERIFICATION. Only use it when an upstream layer verifies the signature. Not recommended in production
editSafetyDelayMs	No	Minimum age (ms) a freshly posted message must reach before its first edit is sent, so receiving clients have stored the original the edit targets. Defaults to 5000; 0 disables the wait
sendReadReceipts	No	Send a read receipt for each delivered inbound message before handlers run (default true)
userName	No	Bot @handle for mention detection. Auto-detected from X_BOT_USERNAME, otherwise resolved from GET /2/users/me
welcomeMessage	No	Message posted when the bot joins a group. false disables; omitted uses a default that explains @mention-to-reply
verifySignatures	No	Require verifiable signatures on incoming messages (default true). X_VERIFY_SIGNATURES=false opts out
signingKeyVersion	No	Signing key version override. Normally fetched during initialize(). Auto-detected from X_SIGNING_KEY_VERSION
apiBaseUrl	No	Base URL for media REST calls (defaults to https://api.x.com)
apiHeaders	No	Extra headers on every X API request. Unless it includes a User-Agent, the adapter prepends chat-sdk-xchat/<version> to the client's default so Chat SDK traffic is identifiable
logger	No	Logger instance (defaults to ConsoleLogger("info"))

*Required at runtime — either via config or environment variable.

Environment variables

XCHAT_BOT_TOKEN=xcbot_...          # OAuth2 user token for the bot account
XCHAT_PIN=...                      # Juicebox PIN for key unlock
X_CONSUMER_SECRET=...              # App secret (CRC + webhook signature verification)
X_BOT_USERNAME=...                 # Optional @handle override; resolved from /2/users/me otherwise
X_VERIFY_SIGNATURES=true           # Optional; set false to accept unverifiable messages

Features

Messaging

Feature	Supported
Post message	Yes (encrypted + signed)
Group replies	Yes (quoted reply to the triggering message)
Edit message	Yes (encrypted edit event targeting the original's sequence id; only the bot's own text messages). The first edit of a fresh message is held until the message is editSafetyDelayMs old, so receiving clients have stored the original before the edit arrives
Delete message	Yes (delete-for-all: a locally signed delete action removes the message for every participant; only the bot's own messages)
Streaming	Limited (message edits work, but the first edit is age-gated by editSafetyDelayMs, so rapid token-by-token updates are coarse)
Read receipts	Yes (sent for each delivered inbound message unless sendReadReceipts: false; falls back to the latest conversation event)
TTL propagation	Yes (replies inherit the inbound message's disappearing-message TTL)

Rich content

Feature	Supported
Markdown	No client-side rendering — outgoing markdown appears literally as plain text (tables are the exception, see below)
URL / @mention entities	Yes (detected in outgoing text, rendered as tappable links and mention pills)
Post cards	Yes (first x.com/.../status/... URL in outgoing text auto-attaches)
Media out	Yes (files encrypt-streamed and uploaded via the 3-step media upload flow)
Media in	Yes (attachments exposed with lazy download + decrypt via fetchData())
Cards / buttons / modals	No (X Chat has no interactive message surface)
Tables	ASCII code blocks

Conversations

Feature	Supported
Mentions	Yes (structured mention entities, swipe-reply-to-bot, plain-text @handle fallback)
Add / remove reactions	Yes
Incoming reactions	Yes (routed to onReaction)
Typing indicator	Yes (keep-alive re-sent every 3s while a handler runs)
DMs	Yes
Group chats	Yes (g-prefixed conversation IDs)
Group join welcome	Yes (configurable via welcomeMessage)
Open DM	Yes (openDM(userId) reuses an existing conversation or runs a fresh key exchange; requires the recipient to have encrypted chat set up)

Message history

Feature	Supported
Fetch messages	Yes (GET /2/chat/conversations/{id}/events, batch-decrypted)
Fetch thread info	Yes
List threads	No

Mention behavior in groups

chat.onNewMention(handler) fires for group messages that mention the bot. A group message counts as a mention when:

1. Its rich-text mention entities include the bot's @handle or user ID — the authoritative signal, or
2. It is a swipe-reply to one of the bot's own messages, or
3. Its plain text contains @handle (fallback when no entities are present).

To reply to every group message instead, register a catch-all chat.onNewMessage(/.+/, handler).

Thread ID format

xchat:{conversationId}

• 1:1 conversations: xchat:1234567890-9876543210 (both participant IDs)
• Group conversations: xchat:g123456789 (opaque g-prefixed ID)

REST paths use the other participant's user ID for 1:1 conversations and the g... ID for groups; the adapter converts automatically.

Encryption notes

• Conversation keys arrive via KeyChange events and are cached per conversation and key version. When a key is missing at send time, the adapter fetches recent conversation events to extract one.
• Incoming message signatures are verified against participants' signing keys (fetched from GET /2/users/{id}/public_keys and cached). With verifySignatures: true (the default), unverifiable messages are dropped.
• Media is encrypted separately from message text using streaming encryption; the message carries a media_hash_key reference that the recipient uses to download and decrypt the blob.

Troubleshooting

CRC validation failing

• Confirm X_CONSUMER_SECRET matches your app's consumer secret in the developer portal
• The response must be {"response_token": "sha256=<base64 HMAC>"} for GET requests

Events not arriving

• Verify the subscription exists: GET /2/activity/subscriptions with the app bearer token
• Confirm the subscription's webhook_id points at a webhook whose URL is your live endpoint
• Subscription changes can take a few minutes to provision

"Invalid signature" on webhook POSTs

• The adapter verifies x-twitter-webhooks-signature with HMAC-SHA256 keyed by X_CONSUMER_SECRET — make sure it is the same app that owns the webhook

Messages decrypt but the bot never replies in groups

• Group replies require a mention — check that the sender actually @mentioned the bot's handle (the one from /2/users/me), or use { allGroupMessages: true }

"No conversation key" when posting

• The bot can only send in conversations where a conversation key has been shared with it (i.e., someone messaged the bot first, added it to the group, or the bot opened the conversation itself via openDM(userId)). openDM requires the recipient to have encrypted chat set up, and the server requires the recipient to trust the bot (e.g. follow it) before the first message is accepted.

AI Coding Agents

If you use an AI coding agent such as OpenAI Codex, Claude Code, or Cursor, install the Chat SDK skill so it knows the SDK APIs, adapter patterns, and project conventions before writing code.

npx skills add vercel/chat

The skill references bundled documentation in node_modules/chat/docs, plus adapter guides and starter templates in the published package.

You can also install the Vercel Plugin for a broader agent toolkit: it includes the Chat SDK skill alongside specialist agents, agent slash commands, and more:

npx plugins add vercel/vercel-plugin

The plugin is optional; the skill alone is enough to build with Chat SDK.

For agent-readable documentation, see chat-sdk.dev/llms.txt (page index) or chat-sdk.dev/llms-full.txt (full text).

License

MIT