Cursor AI releases SDK for creating and improving AI agents
Cursor AI has released an SDK for developers that allows them to create and manage AI agents directly through code. Agents can now be defined using Markdown for prompts and TypeScript for tools, which ensures a high degree of development structure.
Developers gain flexible choice regarding deployment location: they can use Cursor's cloud infrastructure or run agents on their own hardware. A key feature is the built-in evaluation system, which allows measuring the effectiveness of every change and systematically improving agent performance over time.
This significantly simplifies the process of creating complex, iteratively improving AI systems for a wide range of tasks.
Why it matters
- —Provides developers with a structured and code-based approach to creating AI agents.
- —Built-in evaluation tools allow measuring and systematically improving agent performance.
- —Deployment flexibility (cloud or local) expands usage possibilities.
Key facts
- The agent SDK is now available from cursor_ai.
- Agents are defined as code, using Markdown for prompts and TypeScript for tools.
- Supports deployment both on Cursor hosting and locally.
- Includes a built-in evaluation system for monitoring and improving agents.
The full text is in the original source. Here we provide a brief summary and key facts.