Devin can now build internal web apps your team opens with a link
Need an admin panel, a dashboard, or an approval queue, but hate to spend developer time on it? Devin now offers Apps: you describe the tool in words, and the agent writes it and publishes it on its cloud machine. The app gets a permanent link and can be accessed only after signing in to Devin. It sees your MCPs, integrations, and codebase. The machine goes to sleep when idle, and you pay only for the time it is awake. For internal tools, that makes it a cheap replacement for spreadsheets and homemade scripts.
AI-processed from Windsurf Blog; edited by Hamidun News
Almost every team has tasks that are too small for full-scale development but too awkward for a shared spreadsheet: an intake form, a review queue, a project status page, a dashboard for support tickets. Usually such tools either wait for years in the backlog or turn into a fragile spreadsheet that nobody wants to update. Devin's developers propose a different path: Devin Apps, private web apps that the AI agent builds and hosts itself on its cloud machine.
How it works
Users open the Apps section in the sidebar and describe what the app should do. They can start from a ready-made template, such as a metrics dashboard, a project status page, or a review queue. After clicking Build with Devin, the agent starts a session, asks clarifying questions in its first message, then builds the app and deploys it to the web. While the first version is being built, the Apps page shows the status Building, and the app's address is visible right away.
There is a second path. If Devin has already shown a browser preview in a regular session, you can tell it: publish this as an app.
The key practical difference is that the app runs directly on Devin's machine. As a result, it has access to the MCP servers, integrations, and codebase already connected. You can build the app on your own internal stack and components. Any stack that serves HTTP on a port will work. The database can also live on the same machine, for example SQLite or Postgres. In addition, the app can call Devin's own API, so you can build custom tools on top of the agent.
Access, updates, and cost
Every app gets a stable link. There are no public apps: to open any of them, you must sign in to Devin. Using the Share button, you can restrict access to people on a list, open it to the whole organization, or to the entire enterprise. Organization administrators can pin an app for everyone.
To change something, click Edit in session and explain the changes to Devin. The agent builds a new version, and the address does not change. All versions are listed in the Deployments section. If the latest build fails to start, the previous one continues to respond. You can also automate updates: for example, pull fresh data on a schedule or redeploy the app on every branch change.
Under the hood are memory snapshots of the machine. Thanks to them, Devin goes to sleep between tasks, and the app wakes up when someone opens the link. The first load after sleep may take several seconds. You pay for the time the machine is awake. For rarely used internal dashboards, this is cheaper than a constantly running server.
Where it fits and what the limits are
The company names three scenarios. The first is internal tools: admin panels, intake forms, approval queues, and small utilities instead of a shared spreadsheet. The second is dashboards, where Devin pulls data from your sources itself and refreshes the page on a schedule. The third is beta and staging versions: the agent can run your real stack on its machine, with frontend, backend, database, and workers, and colleagues can open a copy of the product via a link.
The limitation is stated plainly: the app runs on a single machine. That suits internal tools and dashboards but not high-load production services. Also, the possible delay after waking up means you should not expect instant responses.
For product managers and team leads, the takeaway is simple. If you have accumulated small internal tools that never made it into development, Apps let you test the idea in a single session without assigning an engineer. For developers, it is a way to quickly show colleagues a working beta. But critical customer-facing services should still be kept on regular infrastructure, and before sharing the link, check who exactly has access: the organization or the entire enterprise.
- Nov 21, 2026FDE job postings grew 4.6x in a year: what a forward deployed engineer actually does at the client's site
- Nov 21, 2026Sakana AI Opens Applied Research Lab for Market and Economic Modeling
- Nov 21, 2026Meta Is Reportedly Preparing the $300 Muse Charm AI Tamagotchi With a Year of Verizon 5G
- Nov 21, 2026Dreamteam Raises $53M for a CRM That Fills Itself from Email and Calls
Key Questions Answered
How it works?
Users open the Apps section in the sidebar and describe what the app should do. They can start from a ready-made template, such as a metrics dashboard, a project status page, or a review queue. After clicking Build with Devin, the agent starts a session, asks clarifying questions in its first message, then builds the app and deploys it to the web. While the first version is being built, the Apps page shows the status Building, and the app's address is visible right away. There is a second path.
What are the key details on Access, updates, and cost?
Every app gets a stable link. There are no public apps: to open any of them, you must sign in to Devin. Using the Share button, you can restrict access to people on a list, open it to the whole organization, or to the entire enterprise. Organization administrators can pin an app for everyone. To change something, click Edit in session and explain the changes to Devin. The agent builds a new version, and the address does not change. All versions are listed in the Deployments section.
Where it fits and what the limits are?
The company names three scenarios. The first is internal tools: admin panels, intake forms, approval queues, and small utilities instead of a shared spreadsheet. The second is dashboards, where Devin pulls data from your sources itself and refreshes the page on a schedule. The third is beta and staging versions: the agent can run your real stack on its machine, with frontend, backend, database, and workers, and colleagues can open a copy of the product via a link.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.