Coder brings Cursor cloud agents into customer-controlled workspaces

Coder has launched Agent Relay for Cursor in private preview with design partners, letting Cursor cloud agents execute inside workspaces on infrastructure controlled by the customer. The pitch is straightforward: keep the coding agent, but move its execution into the organisation’s own security perimeter.

Cursor Watch analysis

What happened

Coder says Cursor agents can run in customer cloud, virtual private cloud or on-premises environments. Developers keep using Cursor across its apps, web and mobile, while agent tool calls execute in Coder-managed workspaces on the customer’s network. The integration is in private preview, not general availability.

Coder says platform teams can set network egress policies, define approved models and data sources, and provision workspaces centrally. It describes agent environments as sandboxed, temporary and limited to a single task, with each run producing a log of what the agent accessed, executed, changed or was blocked from doing. Those are the companies’ described controls, not independent security findings. Read Computer Weekly’s report.

Why it matters

For organisations that cannot send sensitive code or internal services to a vendor-hosted execution environment, where an agent runs can be as important as which model it uses. Coder’s approach aims to let teams adopt Cursor’s cloud agents while keeping execution, network rules and audit records under their own control.

That could make agentic coding more practical in regulated settings, but a private preview is a first step, not proof that compliance teams will sign off. The usefulness will depend on how the controls work in real deployments and how much operational effort they add.

Our read

This tackles a real deployment hurdle rather than asking security teams to admire a new model and look the other way. The strongest part of the proposition is that permissions and records are handled at the environment level, not left entirely to an agent’s good intentions. For now, treat it as a promising route to test, not a security guarantee.

What to watch

• When Agent Relay moves beyond private preview and which customers can use it.
• What controls and audit records are available in actual deployments.
• Whether the setup works smoothly with organisations’ existing security and compliance requirements.

Discussion spark: Would running coding agents inside infrastructure your organisation controls be enough to satisfy its security team, or should sensitive code stay away from cloud agents altogether?

Independent WittyWires tracker for public updates about Cursor. Not affiliated with or endorsed by Cursor; this is not an official account.