AGENTIC AI

INTELLIGENT TECHNOLOGY

The integration expands Cloudflare’ s support for AI agent platforms, which also includes Devin Outposts and Claude Managed Agents. It is designed to address growing enterprise demand for greater control over where AI agent workloads run and how they access source code, systems and secrets.

With Cursor Self-Hosted Machines, Cursor continues to manage the agent loop, including inference, planning and orchestration, while coding tasks are executed on infrastructure selected by the customer. Cloudflare Sandboxes can now provide that execution environment within the customer’ s Cloudflare account.

Developers continue to start and manage agents through Cursor’ s desktop, web or mobile applications, with Cursor routing tasks and returning results. However, tool calls involving terminal, filesystem and browser actions can run within isolated Cloudflare sandbox environments.

This approach is intended to support organisations with strict requirements governing where code, build caches and secrets are stored and processed.

“ Developers want powerful AI tools that fit naturally into their workflows, and enterprises need those tools to run in environments they control,” said Dane Knecht, Chief Technology Officer at Cloudflare.“ Cloudflare Sandboxes gives teams the freedom to use the AI tools they prefer while giving organisations a secure, isolated place to run agent work. Bringing Cursor Cloud Agents to Cloudflare

Sandboxes is another step toward making Cloudflare the execution layer for the next generation of agentic applications.”

Cursor Cloud Agents running through self-hosted machines use an outbound connectivity model. A Cursor worker runs the Cursor CLI and establishes a long-lived outbound HTTPS connection to Cursor’ s backend, through which agent tool calls are sent. This means Cursor does not need to establish an inbound connection to the customer’ s network.

The integration supports individual developers as well as enterprise teams.

Dane Knecht, Chief Technology Officer at Cloudflare

Individual users can connect a worker through Cursor’ s My Machines capability, while enterprises can use self-hosted worker pools as routing targets for agent tasks.

These pools can be configured for different execution environments, allowing organisations to manage capacity according to demand. Pool orchestration can monitor demand, start worker capacity when required and release it when sessions have finished.

During self-hosted operation, repositories, build caches and secrets remain on customer-controlled machines. However, file chunks accessed by the model during inference and Cloud Agent artefacts, including screenshots, videos and log references, are uploaded so they can be displayed in pull requests and dashboards.

Teams can also use the Cloud Agents API to integrate self-hosted machine status and pool routing into their own systems.

The collaboration reflects a broader shift in enterprise AI deployment towards models that combine familiar developer tools with greater organisational control over the infrastructure on which autonomous agents operate. Cloudflare said its Sandboxes provide an isolated execution layer that can support this approach across multiple AI agent platforms. • www. intelligentcio. com

INTELLIGENT CIO MIDDLE EAST

31