Cursor brings self-hosted cloud agents inside enterprise networks

What Cursor said on X

On March 25, 2026, Cursor said its cloud agents can now run on customer infrastructure while preserving the same agent harness and overall workflow experience. The message is straightforward but significant: enterprises no longer need to choose between managed agent ergonomics and keeping code execution inside their own network boundary.

That is a meaningful shift for coding agents because the most useful workflows usually need more than model access. They need terminals, browsers, test environments, internal caches, package mirrors, and network reach into company systems. If those dependencies stay outside the agent runtime, autonomy breaks down quickly.

What Cursor's product post adds

Cursor says self-hosted cloud agents are now generally available. The company describes its agents as isolated virtual machines with a terminal, browser, and full desktop that can clone repositories, set up development environments, write and test code, push changes for review, and continue running while the developer is offline.

The self-hosted variant keeps the codebase, tool execution, and build artifacts inside the customer's environment. Cursor also says these agents can access internal caches, dependencies, and network endpoints in the same way an engineer or service account would. The post positions that as the missing piece for regulated teams or organizations with mature internal environments that cannot move code or secrets outside approved infrastructure.

Cursor also frames self-hosting as a full product surface rather than a reduced-security mode. The blog says teams still get isolated remote environments, multi-model support including Composer 2 and other frontier models, plus plugins that can extend agents with skills, MCPs, subagents, rules, and hooks. Customer examples from Brex, Money Forward, and Notion are used to argue that the model is practical in larger organizations, including one case aimed at enabling nearly 1,000 engineers to create pull requests from Slack.

Why this matters

The larger implication is that agent adoption in enterprise software may now hinge less on raw coding quality and more on deployment topology. Many companies can already imagine useful autonomous coding loops, but security, compliance, and network isolation requirements block rollout. Self-hosted agents turn that blocker into an infrastructure choice rather than a product dead end.

For the broader AI tools market, this is another sign that enterprise agent products are converging on a common requirement: model intelligence is necessary, but operational trust decides whether adoption scales. Cursor is betting that orchestration, isolation, and deep integration with internal environments are now as strategic as the underlying coding model.

Sources: Cursor X post · Cursor product post