feat: add @chat-adapter/web — browser chat UI for chat-sdk bots

[bensabic]

add @chat-adapter/web

A new platform adapter that lets a chat-sdk bot serve a browser chat UI alongside Slack, Teams, Discord, etc. The same bot.onDirectMessage(...) handler fires for every platform — including streamed replies via thread.post(stream).

The adapter speaks the AI SDK UI message stream protocol, so @ai-sdk/react's useChat and the ai-elements component library work out of the box. No client-side glue.

// server
const bot = new Chat({
  userName: "mybot",
  adapters: {
    web: createWebAdapter({
      userName: "mybot",
      getUser: (req) => ({ id: getUserIdFromCookie(req) }),
    }),
  },
  state: createMemoryState(),
});
export const POST = bot.webhooks.web;

// client
import { useChat } from "@chat-adapter/web/react";
const { messages, sendMessage, status } = useChat();

What's in the package

@chat-adapter/web ships two subpath exports:

• @chat-adapter/web — server-side createWebAdapter({ userName, getUser }) that produces an Adapter for the Chat constructor. Handles webhook parsing, user resolution, and streaming the response.
• @chat-adapter/web/react — thin client wrapper exposing useChat() preconfigured with DefaultChatTransport. Re-exports UIMessage and UseChatHelpers types.

How the pieces fit

browser (useChat)                   server                          chat-sdk
─────────────────                   ──────                          ────────
sendMessage ────POST /api/chat────► bot.webhooks.web
                                    └─ WebAdapter.handleWebhook
                                       ├─ getUser(req) ───── 401 if null
                                       ├─ build Message from last UIMessage
                                       ├─ createUIMessageStream({execute})
                                       │   └─ webRequestContext.run(ctx, ──────► chat.processMessage
                                       │       () => chat.processMessage(...))      └─ onDirectMessage(thread, msg)
                                       │                                                  └─ thread.post(...) ──┐
                                       │                                                                        │
                                       │   postMessage / stream read AsyncLocalStorage                          │
                                       │   ctx.writer to emit text-start/delta/end ◄────────────────────────────┘
                                       └─ createUIMessageStreamResponse  (SSE; x-vercel-ai-ui-message-stream: v1)

isDM: true routes every web message through onDirectMessage. channelIdFromThreadId === threadId keeps channel.messages scoped per useChat conversation. persistMessageHistory: true (default) backfills thread.messages from state since web has no platform history API.

Notable design choices

getUser is the security boundary

Web requests come straight from a browser, so unlike the platform adapters there's no signature to verify. getUser is the auth boundary: returning null → HTTP 401 and no handler runs. The adapter's job is to plug into whatever the host app already uses (NextAuth, Clerk, custom session cookies). User ids that contain the reserved : delimiter are rejected with HTTP 400 to keep the thread-id round-trip clean.

Native streaming end-to-end

thread.post accepts AsyncIterable<string | StreamChunk> and pumps deltas straight onto the SSE response body — no edit loop, no rate-limit concerns. Plays nicely with the AI SDK's streamText. request.signal is forwarded into the stream loop, so useChat's stop() short-circuits the iterator on the server side. task_update / plan_update chunks have no native v1 representation in the UI message stream and are dropped silently.

chat.processMessage returns a Promise

The Web adapter response body is the user handler's output stream, so we need to surface handler errors to the client. Chat.processMessage previously returned void; it now returns Promise<void> that rejects on handler failure. Existing webhook adapters using options.waitUntil are unchanged: the SDK still tracks the work with errors swallowed (logged) so platforms don't retry on handler bugs.

Message persistence defaults to true

Web has no platform-side history API, so thread.messages / channel.messages are only populated through the configured state adapter's message history cache. Defaulting to true matches the typical use case; opt out only if your handler re-derives history from the request body's messages[] itself.

v1 scope

In: text + markdown, native streaming, DM-style routing (isDM: true), persisted message history, abort propagation via request.signal, useChat-compatible useChat hook.

Out (deferred to v2): cards/JSX rendering, reactions, modals, file uploads, edit/delete, multi-tab proactive push.

Example app

examples/nextjs-chat gains:

• /chat page using useChat with ai-elements components
• /api/chat route delegating to bot.webhooks.web with Next.js after() for tracking
• Web adapter wired into the existing bot instance — same handlers fire from Slack and from the browser

Tests

• packages/adapter-web/src/index.test.ts — 18 tests covering construction, thread-id encoding, input validation (400/401 paths), end-to-end handler dispatch (onDirectMessage routing, async-iterable streaming, error propagation), and direct stream() coverage (abort short-circuit, dropping non-text task_update/plan_update chunks, SentMessage.id matches the streamed text-* event id).
• packages/chat/src/chat.test.ts — new tests for the awaitable processMessage contract (resolves on success, rejects with the original error on handler failure, waitUntil tracks both).

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
chat	Ready	Preview, Comment, Open in v0	May 5, 2026 9:50pm
chat-sdk-nextjs-chat	Ready	Preview, Comment, Open in v0	May 5, 2026 9:50pm

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	npm/​@​ai-sdk/​react@​3.0.176
	npm/​ai@​6.0.6 ⏵ 6.0.174	-7

View full report

[gr2m]

Code changes look good to me! Very cool!