Cognition, the maker of the Devin coding agent, now lets Devin ship finished software instead of only code. On November 13 the company introduced Devin Apps, private web applications that Devin builds from a description and hosts on its own cloud computer. A team asks for the tool it needs, and Devin hands back a link.

The change matters because it alters what the agent delivers. A pull request still needs someone to deploy it. A hosted app is something a finance lead or support manager can open the same afternoon. Cognition is aiming its agent at the backlog of small internal tools that engineering teams rarely get around to building.

Creating an app starts in the Apps section of the Devin sidebar. The user describes the app or chooses a starter prompt (a metrics dashboard, a status page for a project, or a review queue), then clicks Build with Devin. Devin opens a session, asks its questions up front, builds the app, and lists it with a URL. The app carries a Building label until its first version goes online. A user can also request an app mid-session, including by publishing a browser preview Devin has already shown.

Because the apps run on Devin’s own machine, they can reach a team’s existing MCPs, integrations, and codebase, according to Cognition’s announcement. Teams can build with their internal tech stack and components. That access is the real selling point, since an approval queue or dashboard is only useful if it reads the company’s own data.

Access control is strict. Every app gets a stable URL, everyone who opens one must sign in to Devin, and apps are never public. The Share button limits an app to named people, the whole organization, or the whole enterprise. Admins can pin an app for everyone. That design rules out customer-facing use, which looks intentional. It also means each viewer needs a Devin login, a cost the announcement does not discuss.

Updates keep the address. Someone clicks Edit in session and tells Devin what to change, and each change becomes a new version listed under Deployments. Should a fresh version crash on launch, the prior one stays online. An automation can also pull new data at set times or redeploy whenever a branch moves.

The runtime rests on memory snapshots. The machine hibernates when nobody is using the app and revives on the next visit, so a visitor arriving after a nap waits several seconds. Billing follows the time the machine is awake. Any stack that answers HTTP on a port is allowed, SQLite or Postgres can sit on the same machine, and an app can call the Devin API to build tools on top of Devin itself.

Cognition states the limits plainly. An app runs on one machine, which suits internal tools and dashboards but not high-traffic production services. Besides admin panels, request forms, and spreadsheet replacements, the company points to beta and staging copies of a product, with the frontend, backend, database, and workers all running on Devin’s machine and teammates opening the result from a link.

The announcement gives no hourly rate for awake time. A one-dashboard pilot is the cheapest way to learn whether a spreadsheet replacement stays cheap once the whole team keeps it open all day.

Cognition, the maker of Devin, announced Devin Apps on the devin.ai blog on November 13, 2026.