Today we’re introducing Devin Outposts: run Devin Cloud sessions on any machine. A GPU box in your lab, a VM inside your private network, a Kubernetes cluster next to your internal services, or even a Mac mini on your desk.
How it works
Outposts are named queues that connect Devin Cloud sessions to a group of machines you operate. Create a new Outpost on Devin Cloud, and add an Outpost worker on your machine with the Devin CLI.
Once an outpost is defined, you will be able to select it as your runtime environment on Devin Cloud. Simply select your outpost and start a new Devin session.
The session enters the outpost queue; one of your workers claims it (claims are atomic, so exactly one wins) and runs it with per-session state.
You don’t need to keep workers running around the clock either: the queue is exposed through an API that tells your infrastructure exactly which sessions are waiting at any moment, so an orchestrator can spin machines up on demand and tear them down when the session ends. We include a reference implementation for Kubernetes at devin-outpost-k8s. You can also orchestrate Devin Outposts using popular platforms like Modal, Cloudflare, Daytona, NVIDIA Brev, NVIDIA OpenShell, Namespace, E2B and more.
You can't really build iOS apps without a Mac. So we gave Devin one. With Devin Outposts on Namespace Devboxes, Devin now has a first-class macOS environment, powered by Apple's M5 chip. And with computer use, Devin can autonomously build, run, and test iOS apps end to end.
Devin Outposts running on Cloudflare give every session an isolated sandbox, with all traffic flowing through customizable proxies and private connectivity to internal services. No VPN and no public exposure needed.
With Devin Outposts on Modal, Devin can run on the same infrastructure where you train and serve your models. It can reproduce failures, profile the fix, and scale back to zero, on the same hardware you use in production.
Combining Devin Outposts with the NVIDIA OpenShell runtime gives every Devin session a sandbox. Whether you're running on a single VM or a GPU cluster, an OpenShell runtime on your Outpost workers enables agents to run autonomously via a more secure and government environment.
Outposts on Daytona let Devin sessions hit the ground running, serving Linux and Windows sandboxes under 90ms from a snapshot, with repos, dependencies, and toolchains already in place.
At E2B, we've enjoyed kicking work over to Devin, but occasionally wanted more from his computer. With Outposts, Devin can arm himself with our industry leading Agent machines at any CPU/RAM config he wants and sub-second starts. Can't wait to see what he does with the extra horsepower and access to our customer's private clouds.
How Outpost sessions are different
An Outpost session is still a first-class Devin session: you start it, message it, and watch it work from Devin Cloud exactly like any other, and everything from skills and knowledge to secrets, MCP, and the live desktop view works unchanged. What changes is the ground it stands on. The entire execution now runs on your machine, as an unprivileged user, with your repos, toolchains, GPU drivers, and internal services exactly as they exist on the box.
Outposts unlock a new class of work that Devin can now do with you:
- Debug a failed training run where it failed. Start a session on the GPU box that ran the job. Devin can tail the logs, inspect checkpoints and GPU state, reproduce the crash with the real drivers and the real dataset, and relaunch the run with a fix.
- Work against services only your network can see. Run a worker on a VM inside your VPC and Devin can chase the bug that only reproduces against your staging database, exercise internal APIs end to end, or operate tooling that will never be exposed.
- Ship on new platforms. A Mac mini on your desk becomes a Devin-ready macOS runner: Xcode builds, Apple-silicon tests, and signing with your local toolchain. You can now use Devin to build apps for Apple platforms end-to-end.
When to use Devin Outposts
As we explained in What We Learned Building Cloud Agents, running cloud agents securely and reliably at scale requires significant infrastructure and operational expertise. For most customers, we recommend Devin’s managed hosting options, which provide built-in orchestration, isolation, persistence, scaling, and lifecycle management.
| Option | When to use it | Ownership |
|---|---|---|
| Multi-tenant hosting | The default for most organizations, especially when you want quick deployment and your development resources are publicly accessible. | Cognition manages the infrastructure and orchestration. |
| Dedicated Tenant | Recommended for security-conscious organizations that need stronger tenant isolation, private connectivity, and Cognition-managed security and orchestration. This includes resources accessed through AWS PrivateLink, IPSec, or supported VPN configurations. | Cognition manages a customer-isolated deployment and its orchestration. |
| Devin Outposts | When Devin must run directly on your machines or network and your team already has substantial experience operating remote development VMs securely and reliably at scale. This is generally best suited to tech-native organizations or small, trusted teams. | Your team operates worker capacity, scaling, availability, machine lifecycle, security controls, and orchestration. The operational complexity is comparable to a VPC deployment. |
Run Devin where your work lives
Some work can’t come to the cloud, so we’re bringing Devin to it. Outposts give you Devin’s full autonomy on your own hardware, inside your own network, under your own policies, securely.
Check out the docs and create your first Outpost in seconds.