diff --git a/docs/faq.md b/docs/faq.md
new file mode 100644
index 0000000000..8fb8342cfa
--- /dev/null
+++ b/docs/faq.md
@@ -0,0 +1,76 @@
+# Frequently Asked Questions
+
+## How was build 550 chosen?
+
+A mixture of reasons:
+
+* The early HD era is my favourite. In particular, I like the 'clean'-looking
+  tabs in the user interface, and 550 was the last build with those. I also
+  considered 530, but 550 has a better built-in world map viewer (it supports
+  dungeons in addition to the main surface).
+* Availability of the original loader, client, game unpacker and jaggl jars.
+* Availability of the complete set of original client data files.
+* Availability of a large proportion of the location file encryption keys.
+
+## Why does OpenRS2 use Maven instead of Gradle?
+
+Gradle's task-based model is significantly better than Maven's fixed lifecycle
+model.
+
+However, Gradle's flexibility and rate of development has come at a cost of
+worse IDE integration, to the point at which recent versions of IntelliJ IDEA
+now delegate build actions to Gradle by default rather than attempting to
+understand the project structure. This tends to be slower and consume more
+memory than IDEA's built-in build system. While this setting can be changed, I
+think it is a sign of the future of Gradle's IDE integration.
+
+Furthermore, [nar-maven-plugin][nar-maven-plugin] is, at the time of writing,
+significantly better than Gradle's support for building native code. Gradle's
+new C++ plugin simply doesn't provide enough features.
+
+This might be a decision we revisit in the future.
+
+## Why is OpenRS2 licensed under the GNU GPL?
+
+As significant amount of work went into the development of OpenRS2. My aim is
+to encourage community contributions rather than effort being duplicated across
+multiple independent closed-source forks, making a copyleft license desirable.
+
+I also wanted to frustrate commercial use, given OpenRS2 is itself developed
+entirely non-commercially. While the GPL does this to an extent, the AGPL would
+have been more appropriate. However, it would be far more difficult to enforce
+the AGPL than the GPL, disadvantaging honest users who would have otherwise
+obeyed the license.
+
+A small number of modules (`deob-annotations` and the native library
+replacements) are instead licensed under the LGPL, as it needs to be possible
+to legally link these modules with the proprietary client code.
+
+## Why rewrite the client's native libraries?
+
+Again, there are a mixture of reasons:
+
+* Availability of the original native libraries. I struggled to find the
+  original native libraries for 550, except for 32-bit Windows. While Linux and
+  macOS natives are available for nearby revisions, they are not compatible
+  with the 550 client.
+* The original native libraries were not built for 64-bit Linux and macOS.
+  While this was probably not a major problem in 2009, 64-bit architectures are
+  now the norm.
+* Non-x86 architectures like ARM and RISC-V are becoming more popular. If we
+  start seeing a shift away from x86 on desktop machines, the native libraries
+  will need to be built for those architectures.
+* The original macOS jaggl native library is backed by an NSView, which was
+  deprecated in Java 6 and removed in Java 7. Java 7 requires surfaces to be
+  backed by a CALayer instead.
+* I anticipate that at some point in the future the Linux AWT implementation
+  will be ported from X11 to Wayland, which will require porting the jaggl
+  native library from GLX to EGL.
+* The switch away from OpenGL to newer graphics APIs like Metal and Vulkan
+  might eventually necessitate the inclusion of OpenGL to Metal/Vulkan
+  translation layers.
+* I'm concerned about backwards compatibility and bit rot. The original native
+  libraries were compiled 10 years ago, and at some point one of their
+  dependencies might drop backwards binary compatibility.
+
+[nar-maven-plugin]: https://maven-nar.github.io/