I concur with this. Developing BrowserBox locally is useful for many reasons.<p>Stress testing the application by stretching it like a rubber sheet to wrap around as many different operating systems as possible is a useful way to iron out various bugs that affect more than one system but may not have been triggered in an easier development process.<p>Running the application locally is also a way that many people first download and try it, so ensuring a reasonable experience on a laptop is quite important. Iterating on front-end code with a TLS certificate from mkcert provides access to all the JavaScript APIs you’d expect to see on the public internet under TLS.<p>Running the browser back-end on different OS and architecture targets is a good way to control for or “fuzz against” the quirks you might see in interactions between the operating system, file system layout, the behavior of common Linux-style command line utilities, and the creation of different users and running processes as them. Many of these things have slightly different behaviors across different operating systems, and stretching BrowserBox across those various targets has been one of the strong methods for building stability over time.<p>My respect for Bash as the lingua franca of scripting languages has grown immensely during this time, and I’ve felt validated by the Unix-style approach where commonly available tools are combined to handle much of the OS-level work. Adaptations are made as needed for different targets, while a lot of the business logic is handled in Node.js. Essentially, this approach uses Bash and Linux philosophy as the glue that sticks the Node.js application layer to the substrate of the target operating system. I made this choice a long time ago, and I’m very satisfied with it. I increasingly feel that it’s the validated approach because new features requiring interaction with the operating system, such as extensions or audio, have been well-supported by this design choice for building the remote browser isolation application.<p>An alternative approach might be to stick everything into first-class programming languages, seeking a Node library for each requirement and wrapping anything not directly supported in a C binding or wrapper. But I’ve never found that practical. Node is fantastic for the evented part of the application, allowing data to move around quickly. However, there are many touchpoints between the application and the operating system, which are useful to track or leverage. These include benefits like security isolation from user space, permissions and access control, and process isolation. The concept of a multi-user application integrated with a Unix-style multi-user server environment has been advantageous. The abundance of stable, well-established tools that perform many useful tasks in Unix, and targets that can run those tools, has been immensely helpful.<p>On the front end, the philosophy is to stay at the bleeding edge of the latest web features but only to use them once they become generally available across all major browsers—a discipline that is easier to maintain these days as browser vendors more frequently roll out major, useful features. There’s also a policy of keeping top-level dependencies to a minimum. Essentially, the focus is on the Node runtime, some WebRTC and WebSocket libraries, and HTTP server components. Most of the Node-side dependencies are actually sub-dependencies and not top-level. A lot of dependencies are at the operating system level, allowing us to benefit from the security and stability maintained by multiple package ecosystems. I think this is a sound approach.<p>Porting everything to a Windows PowerShell-type environment was a fascinating exercise. For the front end, having virtually no dependencies except some in-house tools fosters faster iteration, reduces the risk of breaking changes from external sources, and minimizes security risks from frequently updated libraries with thousands of users and contributors. Some of the ways we’ve approached security by design and stability by design include adopting a development process that is local-first but local-first across a range of different targets.