Windsurf: Devin Outposts Let Devin Run On Your Own Machines

Devin's Brain Stays In The Cloud, The Work Moves To You

On July 21, 2026, Windsurf's Devin introduced Devin Outposts, a new deployment model that lets Devin Cloud sessions run on machines a developer or organization already operates instead of exclusively on Cognition-managed infrastructure. The architecture keeps a clean split: Devin's agent loop, meaning the inference and planning that drives the agent's decisions, continues to run in Devin's cloud, while all command execution, file edits, and repository access happen on the user's own hardware.

That split matters because it solves a problem that has quietly limited how far AI coding agents could reach into real engineering workflows: plenty of work simply cannot move to someone else's cloud. Debugging a failure that only reproduces on specific hardware, reaching services that live inside a private network, or building for platforms like iOS that require specific machines are all cases where the agent previously had to stop at the edge of the sandbox. Outposts removes that edge.

How Outposts Work

Setting up an Outpost follows a simple flow. A named queue is created inside Devin Cloud, and Outpost workers are deployed onto whatever machines should service that queue using the Devin CLI, whether that is a Mac mini on a desk, a GPU box in a lab, a private VM, or nodes inside a Kubernetes cluster. Once registered, the Outpost shows up as a selectable runtime environment in Devin Cloud alongside standard options like Ubuntu or Windows. Sessions routed to the queue are claimed atomically by whichever worker is available, and the queue itself exposes an API so infrastructure can be scaled up or down on demand.

On the networking side, workers only need outbound HTTPS access to connect to Devin Cloud. There is no requirement for inbound ports, a public IP, or a VPN tunnel, which keeps the security footprint of an Outpost worker close to that of a normal outbound HTTPS client.

Built-In Support For Specialized Hardware

Outposts can be registered with custom hardware profiles, including GPU-equipped machines, large-memory boxes, and specific OS images, so a team can point Devin at exactly the environment a task needs. This makes Outposts especially relevant for workloads like debugging failures on GPU hardware, inspecting GPU state, reproducing training-run issues, and validating fixes on the hardware where the original problem occurred.

Partner Integrations At Launch

Devin Outposts shipped with support from several infrastructure partners, including Modal, Cloudflare, Daytona, NVIDIA (through both Brev and OpenShell), Namespace, and E2B, each offering guides or pre-built integrations for running Outpost workers on their platforms. That breadth of partner support means most teams already using a sandboxing or compute provider have a fairly direct path to standing up their first Outpost.

Who Outposts Are For

Cognition recommends Outposts primarily for organizations with meaningful experience operating remote development infrastructure securely, tech-native companies, or small trusted teams that need direct machine access or connectivity to network-specific resources. For everyday individual use of Devin, nothing changes: Outposts is an additional option for teams that specifically need Devin to reach infrastructure a standard cloud sandbox cannot.