@twilic/ai-sdk

Vercel AI SDK chat transport and recorder helpers that capture runs as Twilic AI sessions.

Package: @twilic/ai-sdk
Peer dependencies: @twilic/core, ai
Depends on: @twilic/ai
Source: github.com/twilic/ai

Install

pnpm add @twilic/ai-sdk @twilic/ai @twilic/core ai

Quick start

import {
  ensureTwilicInit,
  inspectSession,
  sessionFromEvents,
} from "@twilic/ai";
import { TwilicChatTransport } from "@twilic/ai-sdk";

await ensureTwilicInit();

const transport = new TwilicChatTransport({
  api: "/api/chat",
  record: true,
});

const result = await transport.sendMessages({
  id: "chat-1",
  messages: [{ id: "m1", role: "user", parts: [{ type: "text", text: "hi" }] }],
});

// Consume the UI message stream as usual…
await result.stream.cancel();

const events = transport.recorder?.events ?? [];
const session = sessionFromEvents(events, {
  sessionId: "chat-1",
  source: "ai-sdk",
});
console.log(inspectSession(session));

Exports

TwilicChatTransport

Drop-in chat transport compatible with the Vercel AI SDK ChatTransport surface.

interface TwilicChatTransportOptions {
  api?: string;
  fetch?: typeof fetch;
  recorder?: AIRecorder;
  record?: boolean;
  headers?: Record<string, string>;
}

Behavior:

• POSTs chat messages to api (default /api/chat)
• Accepts SSE (text/event-stream) or .twai (application/vnd.twilic.ai+twai) responses
• When record: true (or a custom recorder is passed), appends parsed events to an AIRecorder

createAISDKRecorder(options?)

Convenience wrapper around createAIRecorder with AI SDK defaults.

Related

• Twilic AI overview
• @twilic/ai
• Example — ai-sdk-transport