diff --git a/README.md b/README.md
index 85f84a5..e62c7cf 100644
--- a/README.md
+++ b/README.md
@@ -1 +1,42 @@
 # mima-tools
+
+## File format
+
+All tools share a common file format, which basically just contains
+the MiMa's initial memory state. Its file extension is `.mima`.
+
+A MiMa operates on words of 24 bits, so the file is split up into
+blocks of 3 bytes, written directly one after the other with nothing
+in-between. The bytes within one 3-byte block are ordered from most
+significant to least significant.
+
+The file contains no metadata. Opcodes are the same as specified in
+the lecture. The first block is at address 0. MiMa's execution starts
+at address 0 (i. e. the first block).
+
+## Programs
+
+These programs are just ideas at this point, but will hopefully be
+working one day.
+
+### mima-run
+
+This program can load and run `.mima` files at a specified processor
+speed (given in instructions per second).
+
+I don't know how IO is going to happen on this machine. Maybe I'll
+just print a memory dump once the `HALT` instruction was encountered,
+or maybe I'll use left over instructions to add a brainfuck-like print
+command. Or maybe there is already a solution designed into the MiMa?
+
+### mima-debug
+
+This program is like `mima-run`, but also contains a live debugging
+TUI.
+
+## Conventions
+
+In the source code, the name MiMa is spelled `Mima`. When displayed,
+it is spelled `MiMa`.
+
+Executable names are all lowercase, and words are separated by a `-`.