IDE Cursor launches version 2.0 with multiple improvements. Up to eight agents can now work in parallel on the same codebase, while a new proprietary model runs 4x faster than comparable alternatives. Voice control is also being introduced.
Cursor introduces Composer, the first proprietary agentic coding model. This frontier model is four times faster than equally intelligent alternatives, the company claims. The model can generate and modify code based on natural language.
The new Plan Mode works in the background. Developers can create a plan with one model and execute it with another. Even parallel agents can be used to review multiple plans simultaneously.
Multi-agent support central
In Cursor 2.0, developers can unleash up to eight agents simultaneously on a single prompt. Via git worktrees or external machines, each agent works in its own isolated copy of the codebase. This prevents conflicts between files.
In addition, there is a new editor interface. A sidebar shows all active agents and their plans. This allows developers to keep track of exactly what is happening in their codebase.
Improved user experience
The prompt interface has been given a makeover. Files and folders now appear inline as pills. Copying and pasting prompts with tagged context has also been improved.
Many explicit items have been removed from the context menu, including @Definitions, @Web, @Link, @Recent Changes, and @Linter Errors. The agent can now collect this context independently, without manual addition to the prompt.
Voice Mode allows the agent to be controlled by voice with built-in speech-to-text conversion. Custom submit keywords can be defined in the settings to start the agent.
Browser and sandboxing generally available
The browser functionality that appeared in beta in version 1.7 is now generally available. Enterprise teams get additional support for using the browser in 2.0. The browser can now be embedded in the editor, with powerful tools for selecting elements and sending DOM information to the agent.
Sandboxed terminals are also now fully rolled out for macOS. Shell commands not on the allowlist run in a sandbox with read and write access to the workspace but without internet access. This significantly increases security.
Tip: AWS introduces agentic IDE that brings prototypes to production