Vercel has introduced HarnessAgent in its AI SDK 7 canary release, providing a single API to run established agent harnesses. According to Vercel's announcement, this feature allows developers to write an agent once and flexibly switch between different runtime environments without refactoring the entire codebase.
Diễn biến
HarnessAgent acts as an abstraction layer, normalizing access to complex components above the model call, such as skills, sandboxes, sessions, and permission flows. In the current experimental canary release, the tool supports initial adapters for Claude Code, Codex, and Pi, with more expected to arrive soon.
Notably, each harness runs the agent in a sandboxed workspace, keeping the host environment completely safe. Furthermore, both HarnessAgent.generate() and HarnessAgent.stream() return AI SDK-compatible results, allowing apps already utilizing tools like useChat to integrate it without modifying their frontend code.
Vì sao đáng chú ý
For the developer community building AI applications, managing and switching between different agent frameworks has traditionally been complex and time-consuming. Vercel's new solution significantly mitigates the risk of vendor lock-in, enabling engineers to freely experiment and choose the most optimal framework for specific tasks without worrying about host system security.