Coder launches Agent Relay to run Cursor AI inside regulated enterprises

Coder has announced Coder Agent Relay, a self-hosted execution environment that allows cloud-based coding agents to run entirely within infrastructure controlled by the customer. SpaceXAI, maker of the Cursor coding assistant, is the launch partner: Cursor's agent loop, including inference and planning, continues to run on SpaceXAI's side, while tool calls execute inside Coder workspaces on the enterprise's own network. The product is currently in private preview with design partners.

The collaboration is squarely aimed at regulated industries, including financial services, government and defence, where vendor-hosted AI tooling has been effectively locked out by compliance and data-residency requirements. According to Coder, the constraint has never been developer appetite for AI tools; it has been the deployment model those tools ship with.

What Agent Relay does

Each Coder workspace can spin up a Cursor worker that opens an outbound connection to Cursor's infrastructure. Platform teams provision and scale those workspaces using the same tooling they already use to manage developer environments. Source code, secrets and internal services do not leave the customer's network; egress policy is set once at the platform level and applied uniformly. Agent environments are sandboxed and ephemeral, scoped to a single task, so a prompt-injection attempt that might redirect an agent toward unauthorised resources is blocked at the environment layer rather than relying on the model to refuse. Every run produces an audit log covering what the agent accessed, executed, changed or was blocked from doing.

Rob Whiteley, chief executive of Coder, said: "Enterprises never rejected AI agents. They rejected the deployment model. What has kept this work out of production is everything that comes after the pilot: where the code runs, who has access, what the agent can reach, and what record exists afterward."

Market context

Agentic coding is maturing rapidly, with a growing cluster of tools that can autonomously write, test and refactor code across a codebase. The category is dominated by a handful of well-funded players, and enterprise security teams have consistently cited data-exfiltration risk and auditability gaps as barriers to production rollout. Gartner figures cited in the release project that 80 per cent of enterprise software engineers will need to upskill for generative AI by 2027, a trajectory that puts pressure on security and platform teams to find compliant pathways rather than simply blocking tools outright.

Coder's positioning as an "AI Operating Layer" sits at an interesting intersection of the developer-environment and AI-governance markets. The company reports more than 125,000 GitHub stars and says it already serves Global 2000 enterprises and government agencies with its self-hosted workspace platform. Competitors in the secure, self-hosted developer environment space include JetBrains with its remote development offerings and cloud-native platforms that have added private-network options, though none has yet paired tightly with a major agentic coding tool under a co-branded go-to-market arrangement of this kind.

Regulatory read-across

Regulated enterprises evaluating Agent Relay will want to map it against several overlapping frameworks. UK financial services firms operating under FCA operational-resilience rules and those subject to DORA in the EU face explicit requirements around third-party technology risk and the ability to audit automated actions. US federal agencies and defence contractors work under FedRAMP and CMMC requirements that mandate where code and data can reside and who can access it. The architecture Coder describes, where inference stays with the vendor but execution is isolated within the customer's perimeter, is a model that aligns in principle with those frameworks, though formal certification status was not disclosed in the release.

SpaceXAI is described as the first of several agent providers Coder intends to bring onto the Agent Relay platform, suggesting the company is positioning the relay infrastructure as a multi-vendor layer rather than a single-partner integration. The speed at which additional partners are announced will be one signal of whether the model gains traction beyond the Cursor user base.