diff --git a/docs/internal-structure.md b/docs/internal-structure.md
index e69de29..4a2c622 100644
--- a/docs/internal-structure.md
+++ b/docs/internal-structure.md
@@ -0,0 +1,33 @@
+# The internal structure of RISCemu
+
+## Loading assembly files:
+In order to load an assembly file, you need to instantiate a CPU with the capabilities you want. Loading an assembly 
+file is the done in multiple steps:
+
+
+* An `RiscVInput` is created, this represents the file internally
+* An `RiscVTokenizer` is created by calling `cpu.get_tokenizer()`.
+* The input is tokenized by calling `.tokenize()` on the tokenizer.
+* The tokens can then be converted to an Executable, this will then 
+  hold all the information such as name, sections, symbols, etc.
+  This is done by creating an `ExecutableParser(tk: RiscVTokenizer)`
+  and the calling `parse()`.
+* Now you have a representation of the assembly file that can be loaded
+  into memory by calling `cpu.load(executable)`, this will internally 
+  construct a `LoadedExecutable`, which represents the actual memory
+  regions the executable contains (and some meta information such as
+  symbols).
+* You can load as many executables as you want into memory. If you want
+  to run one, you pass it to `run_loaded(loaded_bin)` method of the cpu.
+
+You shouldn't have to do this manually, as the `riscemu/__main__.py` has all the necessary code.
+
+## Instruction sets
+Each instruction set is in a separate file in `riscemu/instructions/`. All instruction sets have to inherit from the
+`InstructionSet` class that sets up all the relevant helpers and loading code.
+
+Creating a cpu with certain instruction sets is done by passing the CPU constructor a list of instruction set classes:
+```
+cpu = CPU(config, [RV32I, RV32M])
+```
+