Vercel has released AI SDK 7, a major update to its TypeScript toolkit for building AI applications, frameworks and agents across model providers. The company says the SDK, which it says has more than 16 million weekly downloads, now offers more production-focused features for developers building agentic systems.
The release centers on five broad areas: creating agents, running them more reliably, integrating outside agent frameworks, observing agent behavior, and extending beyond text into other media formats. Vercel says the changes are designed to make the SDK more useful in real-world deployments, not just prototypes.
One of the main additions in AI SDK 7 is support for more detailed control over reasoning. Vercel said the SDK now standardizes reasoning settings behind a single option for generateText and streamText, while still leaving room for provider-specific configuration when needed. The company is aiming to make it easier for developers to tune how much effort a model uses without learning different interfaces for each provider.
The update also adds a typed tool context system. According to Vercel, each tool can now define its own schema for the contextual data it needs, such as API keys or configuration values. That context stays limited to the specific tool, which Vercel says helps reduce exposure of information to third-party tools that do not need it.
These changes are part of a broader effort to support tools that may be built and maintained outside a given application or agent. Vercel notes that outside companies increasingly offer APIs and services through tools that agents can call, making tighter context management more important.
AI SDK 7 also introduces features aimed at making agents more dependable in production. Vercel says developers can now use tool approvals, durability through a component called WorkflowAgent, timeouts, and sandbox support. Together, these additions are intended to help control what agents do, preserve progress during longer tasks, and limit risk when agents interact with external systems.
On the observability side, the SDK gains telemetry, Node.js tracing channel support, lifecycle events and performance statistics. Vercel says these features are meant to help developers understand how agents behave, where they spend time, and how they perform in practice.
The company is also broadening compatibility with external agent harnesses. AI SDK 7 is designed to work with systems including Codex, Claude Code, Deep Agents, OpenCode and Pi. That suggests Vercel wants the SDK to serve as a common layer even when teams use different orchestration tools.
Vercel is also pushing AI SDK beyond text-only use cases. The release adds provider-agnostic support for real-time voice and video generation, expanding the toolkit for applications that need richer, multimodal output.
For existing users, Vercel says upgrading from version 6 should be straightforward. It points developers to an automated codemod and a migration skill intended to reduce the amount of manual code changes required during the transition.
The update lands as AI SDK continues to serve as the base layer for Vercel’s own open-source agent framework, eve. With AI SDK 7, the company is signaling a stronger focus on the operational side of agent development, including control, safety, integration and visibility.