Cursor launched a native iOS app in public beta on June 30, giving developers the ability to start coding agents, track their progress, and merge pull requests without touching a laptop.
The app targets a specific frustration: developers who carry their machines everywhere because something always needs their attention. With the iOS release, Cursor separates the act of directing agents from the act of sitting at a desk. Users pick a repository, describe a task using voice or text, and hand it to an agent that runs either locally on their machine via Remote Control or inside Cursor’s cloud infrastructure.
For on-call engineering, the workflow shift is meaningful. An engineer paged during lunch can launch an investigation agent from their phone, receive iOS Live Activities updates on the lock screen, and arrive back at a desk to find a pull request already waiting. The announcement from Cursor describes a similar pattern for customer-reported bugs: a developer away from the office starts an agent to reproduce the issue, inspect the relevant code, and work toward a fix, all before opening a laptop.
The trust question embedded in this workflow is worth naming. Merging a PR from a phone condenses a decision that normally involves reading a diff, checking test results, and reviewing context accumulated across a full session. Cursor’s cloud agents produce artifacts alongside code: demos, screenshots, and logs. But the review surface on a small screen is narrower than a desktop IDE, and the social and organizational norms around what constitutes adequate review before merging have not caught up to what mobile tooling now makes possible.
Cloud agents in Cursor run inside isolated virtual machines with full development environments. They operate asynchronously without waiting for local resources, which lets them iterate longer and accumulate more context before surfacing a result. Cursor describes the handoff model as fluid: a local plan can be promoted to a cloud agent, and a cloud session can be pulled back to a local machine for final testing before merge.
The company is also signaling where this goes next. Cursor says the boundary between local and cloud execution will eventually become invisible to the developer. In the near term, the team is adding repo-less chats for tasks that do not require codebase context, and some early users are already running Cursor with MCPs to query Datadog logs and summarize Slack activity, uses that treat Cursor as a general-purpose operator rather than strictly a coding assistant.
The public beta is available on all paid plans. Cursor is offering 75 percent off on Composer 2.5 runs in the mobile app through July 5, 2026.
For engineering teams currently defining on-call runbooks, the question of whether a phone-initiated agent action counts as an adequate incident response is now a policy decision, not just a tooling one.
Source: Cursor company blog (cursor.com), published June 30, 2026.
