Mini-App Shell

A mini-app shell is a product architecture built around a minimal host frame and a launcher that loads independent task-specific surfaces on demand. The frame stays out of the way after the initial summon. Each surface arrives with its own layout, controls, density and even design tokens when it makes sense. The shell handles authentication, data fetching and context passing but it never imposes a global sidebar or top navigation that follows you from surface to surface. This pattern emerged because the old sidebar model could not keep up with denser screens, command bar navigation and AI generated workflows that change what the user needs every thirty seconds. By 2026 the best products had turned their entire UI into a collection of these mini-apps that users assemble on the fly. The architecture is deliberately liquid. It assumes you will use ten different surfaces in one hour and it makes each one feel purpose built rather than like a view inside a larger app. Power users love it because the chrome tax disappears. The work fills the screen until it does not need to anymore.

It is not just any modular interface or component library. It is not a single page app with clever routing. It is not a dashboard that swaps main content areas while keeping the same rails. Those approaches still enforce one product taxonomy and one set of global controls. A real mini-app shell rejects the idea of a single taxonomy. It treats each job as its own application that can break the rules if breaking them serves the task. If your so-called mini-apps all share the same header logo and footer links then you simply rebuilt the sidebar you were trying to escape. The pattern requires courage to let surfaces diverge. Without that courage you end up with the worst of both worlds. Inconsistent enough to feel sloppy but not different enough to feel optimized.

Raycast set the bar in 2023 with its extension system and has not looked back. Each command opens a surface that behaves like its own app. The GitHub mini-app lets you browse issues, review code and merge PRs without ever seeing the Raycast command bar again until you finish. The weather surface shows radar maps and hourly forecasts in a layout no other surface uses. Developers have built mini-apps for everything from linear ticket creation to figma comment triage to personal CRM surfaces. The shell is so thin that these extensions feel first party. Vercel followed in 2025 by turning their dashboard into a series of project level mini-apps. Opening a deployment log gives you a full screen terminal like experience with its own filtering syntax and keyboard shortcuts that would never make sense in the analytics surface. The two surfaces share your team billing status but nothing else in the UI. Slack Canvas launched as a mini-app in 2024 and immediately changed how teams collaborate inside channels. The canvas surface has its own infinite canvas controls, comment threading that looks like Figma and export options that feel like they came from a dedicated design tool. Notion did the same for their databases by 2026. A database launch now opens a dedicated mini-app complete with its own sidebar just for views and its own formula language editor that takes over the entire window. The parent Notion document tree disappears until you close it. In fintech Revolut rolled out investment tracking as a mini-app in Q1 2025. The surface has advanced charting tools, news integration and trade execution buttons that look nothing like the clean transaction list in the main Revolut app. Cursor AI uses mini-app shells for every generated artifact. A refactor suggestion appears in its own surface with inline accept buttons per change and a chat pane that only talks to that particular diff. Close the surface and the changes merge back into the main editor canvas. Height, Pitch and even Arc browser extensions follow the same philosophy. Each booster or view is its own contained universe.

Choose mini-app shells when your users operate in short intense bursts and when AI is part of the workflow handing off tasks that require different surfaces. They pair perfectly with command bars because the bar becomes the way to launch the next mini-app. They reduce visual fatigue and let each surface optimize completely for its job. Raycast users report staying in flow longer because nothing competes for attention when they are inside a surface. The pattern also makes it trivial to test new features. Ship a new mini-app. If it fails you remove it without touching the rest of the product. This is why startups in 2026 ship faster than enterprises stuck with 2018 style monoliths. Avoid mini-app shells when your users need simultaneous visibility into multiple data types like account managers who live in CRMs all day. They also demand strong onboarding. Without it new users see a blank screen and a command bar and have no idea what to type. Mobile is still unsolved territory for most implementations because cramming these surfaces into a phone screen forces compromises that break the magic. Finally skip them if your team cannot enforce minimum consistency standards. When every mini-app invents new icons for the same actions the product stops feeling like one product and starts feeling like chaos.

The mini-app shell finally admits that modern work is a series of quick decisive strikes not a slow march through a fixed hierarchy of pages.