Vercel Bridges the Gap Between AI Chat and Real-World Phone Calls
Vercel has released a new integration that brings Dial—a platform for programmable phone numbers—directly into its Chat SDK, allowing developers to manage SMS, MMS, iMessage, and voice call transcripts as standard chat threads. Announced on the Vercel changelog, the new vendor-official adapter transforms each phone conversation into a Chat SDK thread, with texts, media, and even inbound voice-call transcripts appearing as messages on that same thread.
This is the first time Vercel has added telephony support at the SDK level, signaling a major shift in how developers can build AI-powered customer service bots, virtual assistants, and communication workflows without stitching together separate telephony and chat APIs.
What the Dial Integration Actually Does
The integration works by assigning each phone conversation a stable, per-conversation thread ID within the Chat SDK. When a user sends an SMS or MMS to a Dial-powered phone number, the message arrives as a standard Chat SDK message object. Media files—photos, videos, audio clips—are handled bidirectionally through the same APIs. When a voice call ends, the call transcript is automatically posted as a message on that thread.
All webhooks from Dial are verified using HMAC signatures, ensuring that only legitimate incoming messages and call events are processed. Replies from the AI bot are sent back via the same thread and message APIs, meaning developers only need to learn one set of interfaces for both chat and phone interactions.
Why This Matters for AI Developers
For AI developers building conversational agents, the biggest bottleneck has always been context continuity. A user might start a conversation via SMS, send a photo, then follow up with a voice call. Previously, these interactions would hit different systems—one for text, one for voice—breaking the conversation history and forcing developers to build complex state-stitching logic.
With Vercel’s Chat SDK + Dial, that complexity disappears. The thread ID remains stable across all modalities. An AI bot can:
- Receive an SMS question, respond via text
- Get a follow-up MMS with a screenshot, stored as a message attachment
- Receive a voice call later, with the transcript appearing as another message in the same thread
- Use the full thread history to understand context before responding
This is especially valuable for industries like healthcare, legal services, and customer support, where a single case often spans multiple communication channels. A patient might send a photo of a rash (MMS), then call to describe symptoms (voice), then receive a follow-up text with appointment details. All of these events are now unified in one thread.
Technical Implementation: What Developers Need to Know
To begin using the integration, developers need to include the Dial adapter from the Vercel Chat SDK package. The adapter handles all the low-level telephony logic, including number provisioning, inbound message routing, and webhook verification. Developers only need to define message handlers and thread state logic.
HMAC verification ensures that webhooks are not spoofed—a critical security requirement for any production system handling personal phone numbers or medical data. The stable per-conversation threading means that even if a user calls from a different number later (e.g., their work line), the conversation history won't be lost, provided the same Dial phone number is used.
Bidirectional media support allows bots to send images, PDFs, or audio files back via MMS or iMessage, enabling rich interactions like sending a receipt after a voice-based order.
Business Impact: From Bottleneck to Unified Experience
For businesses already using Vercel’s AI SDK or Next.js, this integration eliminates the need for a separate telephony API provider and the associated integration costs. The shift to unified threads means that AI agents can maintain consistent personality, memory, and context across SMS, iMessage, and voice calls.
We should expect to see new use cases emerge almost immediately: AI sales agents that can handle a full purchase cycle via text and voice, support bots that accept photo uploads mid-call, and scheduling assistants that confirm appointments over iMessage and then call to remind users.
One potential challenge is voice call latency. While SMS and iMessage are near-instant, real-time voice calls require low-latency transcription. The current integration handles transcripts after the call ends, which is fine for support tickets and record-keeping but not for real-time conversational AI during the call itself. Developers building live voice agents will need to pair this with a separate real-time speech-to-text service.
What This Signals for the AI SDK Ecosystem
Vercel’s move to offer a first-party telephony adapter is a strong signal that the company sees the future of AI not just in chatbots but in omnichannel communication. By making phone calls a first-class data type in the Chat SDK, Vercel is positioning itself as the universal middleware for AI-human interaction, regardless of channel.
In 2025, we saw the AI SDK add support for WhatsApp, Telegram, and Slack adapters. The addition of Dial in 2026 completes the picture for most communication channels. Developers can now build a single AI agent that speaks across web chat, mobile push, social messaging, and the PSTN phone network—all managed through one SDK.
The open question is whether Vercel will add real-time voice streaming support in a future update. For now, the post-call transcript model covers many business use cases, but full duplex voice AI will likely require a separate real-time adapter.
Getting Started
Developers can find the Dial adapter in the Vercel Chat SDK documentation, along with example code for setting up a phone number, handling incoming messages, and sending replies. The adapter is available now for all Vercel users with a Dial account.
For teams already using Vercel’s AI SDK, the integration path is minimal: install the adapter, configure your Dial phone number, and start building. The SDK handles the rest—thread management, message routing, and webhook security.
Related: HuggingFace LeRobot v0.6.0: Sim-to-Real Robotics Gets a CI-First Overhaul for Developers
Source: Vercel Blog. This article was produced with AI assistance and reviewed for accuracy. Editorial standards.
