Vercel launches AI SDK 7: complete platform for agents in TypeScript

On March 5, 2025, Vercel announced the release of AI SDK 7, a major overhaul of its popular AI SDK for TypeScript. According to Vercel's official blog, this version moves beyond the limited focus on model calls and chat to become a complete 'agent platform.' The SDK now covers the development, execution, integration, and observation of agents working with text, audio, real-time, image, and video. All major AI model providers are supported out of the box.

What happened?

AI SDK 7 represents the evolution of an SDK that started as a tool to simplify calls to language models. In its earlier versions (since AI SDK 1.0 in 2023), the focus was on basic chat and text generation primitives. With version 7, Vercel makes a qualitative leap by incorporating a complete agent ecosystem. The SDK now includes primitives for reasoning control, tool and runtime context, file and skills support, MCP (Model Context Protocol) applications, and a terminal interface. Additionally, it introduces a harness layer that allows integrating open-source agents like Claude Code, Codex, Deep Agents, OpenCode, and Pi under a single interface. This interoperability reduces ecosystem fragmentation, a problem that has plagued AI agent development, where each tool uses its own API and formats.

According to Vercel's blog, the SDK has been redesigned to support modalities beyond text: real-time, video, voice, transcription, images, files, embeddings, and structured output. This makes it one of the most complete platforms for building multimodal applications. Real-time and video support is particularly relevant, as it opens the door to virtual assistants with live interaction and multimedia content analysis, areas gaining traction in sectors like customer service, education, and entertainment.

A key technical aspect is the mandatory adoption of Node.js 22 and ESM (ECMAScript Modules). Node.js 22, released in October 2024, includes modern APIs like native fetch and AsyncLocalStorage, which improve performance and traceability. The abandonment of CommonJS (require()) is a disruptive change but necessary to align with JavaScript ecosystem trends. Vercel has provided codemods (automatic code transformations) and a detailed migration guide to ease the transition, but developers will need to manually review semantic changes.

Why is it important?

AI SDK 7 marks a turning point for developers building AI applications in TypeScript. Until now, the SDK focused on model and chat primitives; with this version, Vercel bets on a unified ecosystem where agents are first-class citizens. The inclusion of a harness layer allows integrating open-source agents like Claude Code, Codex, or Pi under a single interface, facilitating interoperability and reducing ecosystem fragmentation. This is comparable to what Kubernetes did for container orchestration: standardizing how agents are deployed and managed.

The requirements for Node.js 22 and ESM are disruptive changes but necessary to leverage modern APIs like native fetch and AsyncLocalStorage, which improve performance and traceability. The community will need to migrate, but Vercel offers codemods and a detailed migration guide to ease the transition. In historical context, this migration recalls the shift from Python 2 to Python 3, though Vercel has been more proactive by providing automated tools.

The market impact is significant. Vercel competes with platforms like LangChain, LlamaIndex, and Hugging Face, which also offer tools for building agents. However, AI SDK 7 differentiates itself through native integration with the Next.js ecosystem and Vercel Edge Functions, enabling agent deployment at the edge with low latency. Additionally, support for MCP (Model Context Protocol) aligns it with the standard proposed by Anthropic, which could facilitate adoption in companies already using Claude.

What consequences will it have?

The adoption of AI SDK 7 will accelerate the development of complex agents in production. New observability capabilities (redesigned telemetry, Node.js tracing channels, lifecycle callbacks) will allow teams to monitor and debug agents with greater precision. The expansion to modalities like video and real-time opens the door to innovative applications in virtual assistants, process automation, and multimedia content. For example, an agent could analyze a live video, transcribe audio, extract key images, and generate a structured summary, all within the same flow.

However, migration will not be trivial. Existing projects will need to update their runtime to Node 22 and convert their imports to ESM. Vercel has released a 'migration skill' for AI agents that automates much of the process, but developers will need to manually review semantic changes. This could cause friction in large teams with legacy codebases. Additionally, the dependency on Node.js 22 could be a hurdle for environments still using older LTS versions (like Node.js 20), although Vercel has already announced that its platform supports Node.js 22 since February 2025.

In the enterprise space, AI SDK 7 could consolidate Vercel's position as a leading platform for AI applications on the frontend. Companies like Netflix, Airbnb, and The Washington Post already use Vercel for their websites; they could now extend its use to AI agents. Improved observability will also facilitate adoption in regulated environments where traceability is critical.

What should readers know?

• Node.js 22 is mandatory: The SDK depends on APIs like native fetch and AsyncLocalStorage that are not available in earlier versions. Teams must plan to update their development and production environments.
• ESM is the only supported format: CommonJS (require()) is deprecated. Projects must configure "type": "module" in package.json or use .mjs files. This affects libraries that depend on CommonJS modules.
• New agent primitives: Includes reasoning control, tool and runtime context, file and skills support, MCP applications, and a terminal interface. This allows building agents that can execute code, read files, and make multi-step decisions.
• Enhanced observability: Redesigned telemetry with @ai-sdk/otel, support for Node.js tracing channels, and per-step performance statistics. Teams can now trace every model call, tool call, and reasoning step.
• Agent harness integration: Claude Code, Codex, Deep Agents, OpenCode, and Pi can be wrapped under the same SDK interface. This allows developers to switch agents without rewriting integration logic.
• Multimodal support: The SDK now handles real-time, video, voice, transcription, images, files, and embeddings. This opens use cases like live voice assistants, video analysis, and multimedia content generation.
• UI and streams: Improvements in UI and streaming make agents more reliable and composable, with support for proper streams and message handling.
• Expanded MCP: The MCP package now offers a richer integration surface for agent tools and application UIs, facilitating connection with external services.

"AI SDK 7 is a major release for building production agents in TypeScript. The SDK has grown from model calls and chat primitives into a broader agent platform for developing, running, integrating, and observing agents across text, audio, realtime, image, and video." — Vercel Blog

In summary, AI SDK 7 represents a qualitative leap for agent development in TypeScript, but requires careful migration. Teams that adopt this version will be able to build more robust and multimodal applications, backed by a unified and observable platform. The decision to migrate will depend on the urgency to leverage new capabilities versus the cost of technical upgrade. For startups and agile teams, the benefit is clear; for organizations with legacy systems, migration will require strategic planning.