diff --git a/Makefile b/Makefile
index 9f46246..a64c3e5 100644
--- a/Makefile
+++ b/Makefile
@@ -19,7 +19,7 @@ tests: $(WASM_FILES)
 
 .PHONY: sieve-vanilla
 sieve-vanilla: $(WASM_FILES)
-	$(PARCEL) --no-hmr -o dist/sieve-vanilla.html benchmarks/sieve-vanilla/index.html
+	$(PARCEL) --no-hmr -o dist/sieve-vanilla.html tests/benchmarks/sieve-vanilla/index.html
 
 wasm: $(WASM_FILES) src/tools/quadruple.wasm.hex
 
@@ -27,7 +27,7 @@ dist/waforth.wasm: src/waforth.wat dist
 	racket -f $< > src/waforth.wat.tmp
 	$(WAT2WASM) $(WAT2WASM_FLAGS) -o $@ src/waforth.wat.tmp
 
-dist/sieve-vanilla.wasm: benchmarks/sieve-vanilla/sieve-vanilla.wat
+dist/sieve-vanilla.wasm: tests/benchmarks/sieve-vanilla/sieve-vanilla.wat
 	$(WAT2WASM) $(WAT2WASM_FLAGS) -o $@ $<
 
 dist:
diff --git a/README.md b/README.md
index 0e79373..750558d 100644
--- a/README.md
+++ b/README.md
@@ -1 +1,104 @@
-Nothing to see here (yet)
+# [WAForth](https://el-tramo.be/waforth): Forth Interpreter+Compiler for WebAssembly
+
+WAForth is a bootstrapping Forth interpreter and compiler for
+[WebAssembly](https://webassembly.org). You can see it in a demo
+[here](https://el-tramo.be/waforth/).
+
+It is (almost) entirely written in WebAssembly and Forth, and the compiler generates
+WebAssembly code on the fly. The only parts for which it relies on external
+(JavaScript) code is the dynamic loader (since WebAssembly [doesn't support JIT
+yet](https://webassembly.org/docs/future-features/#platform-independent-just-in-time-jit-compilation)), and the I/O primitives to read and write a character.
+
+The implementation was influenced by [jonesforth](http://git.annexia.org/?p=jonesforth.git;a=summary),
+and I shamelessly stole the Forth implementation of some of its high-level words.
+
+WAForth is still just an experiment, and doesn't implement all the ANS standard words yet. 
+
+## Install Dependencies
+
+The build uses [Racket](https://racket-lang.org) for processing the WebAssembly code, 
+the [WebAssembly Binary Toolkit](https://github.com/WebAssembly/wabt) for converting it in binary
+format,and [Yarn](https://yarnpkg.com) for managing the dependencies of the shell.
+
+    brew install wabt yarn minimal-racket
+    yarn
+
+
+## Building & Running
+
+To build everything:
+    
+    make
+
+To run the development server:
+
+    make dev-server
+
+## Testing
+
+To run the development server with all the tests:
+
+    make tests
+
+## Design
+
+### The Macro Assembler
+
+The WAForth core is written as [a single module](https://github.com/remko/waforth/blob/master/src/waforth.wat) in WebAssembly's [text format](https://webassembly.github.io/spec/core/text/index.html). The 
+text format isn't really meant for writing code in, so it has no facilities like a real assembler
+(e.g. constant definitions, macro expansion, ...) However, since the text format uses S-expressions,
+you can do some small modifications to make it extensible with Lisp-style macros. 
+
+I added some Racket macros to the module definition, and implemented [a mini
+assembler](https://github.com/remko/waforth/blob/master/src/tools/assembler.rkt)
+to print out the resulting s-expressions in the right format.
+
+The result is something that looks like a standard WebAssembly module, but sprinkled with some
+macros for convenience.
+
+### The Interpreter
+
+The interpreter runs a loop that processes commands, and switches to and from
+compiler mode. 
+
+Contrary to some other Forth systems, this system doesn't use a strict threading system
+for executing code. WebAssembly doesn't allow unstructured jumps, let alone dynamic jumps.
+Instead, each word is implemented as a single WebAssembly function, and the system uses
+calls and indirect calls (see below) to execute words.
+
+
+### The Compiler
+
+While in compile mode for a word, the compiler generates WebAssembly instructions in
+binary format (since there is no assembler infrastructure in the browser). Since WebAssembly
+[doesn't support JIT compilation yet](https://webassembly.org/docs/future-features/#platform-independent-just-in-time-jit-compilation), a finished word is bundled into a separate binary WebAssembly module, and
+sent to the loader, which dynamically loads it and registers it with  a shared 
+[function table](https://webassembly.github.io/spec/core/valid/modules.html#tables) at the
+next offset, which in turn is recorded in the word dictionary. 
+
+Because words reside in different modules, all calls to and from the words need to happen as
+indirect `call_indirect` calls through the shared function table. This of course introduces
+some overhead, although it seems limited.
+
+As WebAssembly doesn't support unstructured jumps, control flow words (`IF/ELSE/THEN`, 
+`LOOP`, `REPEAT`, ...) can't be implemented in terms of more basic words, unlike in jonesforth.
+However, since Forth only requires structured jumps, the compiler can easily be implemented 
+using the loop and branch instructions available in WebAssembly.
+
+
+### The Loader
+
+The loader is a small bit of JavaScript that uses the [WebAssembly JavaScript API](https://webassembly.github.io/spec/js-api/index.html) to dynamically load a compiled word (in the form of a WebAssembly module),
+and add it to the shared function table.
+
+### The Shell
+
+The shell is [a JavaScript class](https://github.com/remko/waforth/blob/master/src/shell/WAForth.js) 
+that wraps the WebAssembly module, and loads it in the browser.
+It provides the I/O primitives to the WebAssembly module to read and write characters to a terminal, 
+and externally provides a `run()` function to execute a fragment of Forth code.
+
+To tie everything together into an interactive system, there's a small console-based interface around this shell to type
+Forth code, which you can see in action [here](https://el-tramo.be/waforth/).
+
+![WAForth Console](https://el-tramo.be/waforth/console.gif "WAForth Console")
diff --git a/src/shell/index.css b/src/shell/index.css
index 9811ab3..1a573c6 100644
--- a/src/shell/index.css
+++ b/src/shell/index.css
@@ -40,6 +40,11 @@ body {
   color: #00ff00;
 }
 
+.jqconsole-header a {
+  color: inherit;
+  text-decoration: none;
+}
+
 /* The cursor. */
 .jqconsole-cursor {
     background-color: gray;
diff --git a/src/shell/index.js b/src/shell/index.js
index 50c6d43..b4669a5 100644
--- a/src/shell/index.js
+++ b/src/shell/index.js
@@ -4,6 +4,7 @@ import $ from "jquery";
 window.jQuery = $;
 require("jq-console");
 
+// Copied from https://rosettacode.org/wiki/Sieve_of_Eratosthenes#Forth
 const sieve = `
   : prime? HERE + C@ 0= ;
   : composite! HERE + 1 SWAP C! ;
@@ -30,6 +31,9 @@ const forth = new WAForth();
 
 let jqconsole = $("#console").jqconsole("WAForth\n", "");
 $("#console").hide();
+$(".jqconsole-header").html(
+  "<span><a target='_blank' href='https://github.com/remko/waforth'>WAForth</a>\n</span>"
+);
 let outputBuffer = [];
 forth.onEmit = c => {
   outputBuffer.push(String.fromCharCode(c));
diff --git a/src/waforth.wat b/src/waforth.wat
index 06485db..b5cc330 100644
--- a/src/waforth.wat
+++ b/src/waforth.wat
@@ -1159,67 +1159,45 @@ EOF
       (unreachable)))
 
   ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
-  ;; For benchmarking
+  ;; A sieve with direct calls. Only here for benchmarking
   ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
 
- (func $sieve1_prime
-   (call $here)
-   (call $plus)
-   (call $c-fetch)
-   (call $zero-equals))
-
- (func $sieve1_composite
-   (call $here)
-   (call $plus)
-   (call $push (i32.const 1))
-   (call $swap)
-   (call $c-store))
-
- (func $sieve1 (export "sieve1")
-  (call $here)
-  (call $over)
-  (call $erase)
-  (call $push (i32.const 2))
-  (block $label$1
-   (loop $label$2
-    (call $two-dupe)
-    (call $dupe)
-    (call $star)
-    (call $greater-than)
-    (br_if $label$1 (i32.eqz (call $pop)))
-    (call $dupe)
-    (call $sieve1_prime)
-    (if (i32.ne (call $pop) (i32.const 0))
-     (block
-      (call $two-dupe)
-      (call $dupe)
-      (call $star)
-      (call $beginDo)
-      (block $label$4
-       (loop $label$5
-        (call $i)
-        (call $sieve1_composite)
-        (call $dupe)
-        (br_if $label$4 (call $endDo (call $pop)))
-        (br $label$5)))))
-    (call $one-plus)
-    (br $label$2)))
-  (call $drop)
-  (call $push (i32.const 1))
-  (call $swap)
-  (call $push (i32.const 2))
-  (call $beginDo)
-  (block $label$6
-   (loop $label$7
-    (call $i)
-    (call $sieve1_prime)
-    (if (i32.ne (call $pop) (i32.const 0))
-     (block
-      (call $drop)
-      (call $i)))
-    (br_if $label$6 (call $endDo (i32.const 1)))
-    (br $label$7))))
-  (!def_word "sieve1" "$sieve1")
+;  (func $sieve1_prime
+;    (call $here) (call $plus) (call $c-fetch) (call $zero-equals))
+;
+;  (func $sieve1_composite
+;    (call $here) (call $plus) (call $push (i32.const 1)) (call $swap) (call $c-store))
+;
+;  (func $sieve1 (export "sieve1")
+;   (call $here) (call $over) (call $erase)
+;   (call $push (i32.const 2))
+;   (block $label$1
+;    (loop $label$2
+;     (call $two-dupe) (call $dupe) (call $star) (call $greater-than)
+;     (br_if $label$1 (i32.eqz (call $pop)))
+;     (call $dupe) (call $sieve1_prime)
+;     (if (i32.ne (call $pop) (i32.const 0))
+;      (block
+;       (call $two-dupe) (call $dupe) (call $star)
+;       (call $beginDo)
+;       (block $label$4
+;        (loop $label$5
+;         (call $i) (call $sieve1_composite) (call $dupe)
+;         (br_if $label$4 (call $endDo (call $pop)))
+;         (br $label$5)))))
+;     (call $one-plus)
+;     (br $label$2)))
+;   (call $drop) 
+;   (call $push (i32.const 1)) (call $swap) (call $push (i32.const 2))
+;   (call $beginDo)
+;   (block $label$6
+;    (loop $label$7
+;     (call $i) (call $sieve1_prime)
+;     (if (i32.ne (call $pop) (i32.const 0))
+;      (block (call $drop) (call $i)))
+;     (br_if $label$6 (call $endDo (i32.const 1)))
+;     (br $label$7))))
+;   (!def_word "sieve1" "$sieve1")
   
 
   ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
diff --git a/benchmarks/sieve-vanilla/index.html b/tests/benchmarks/sieve-vanilla/index.html
similarity index 100%
rename from benchmarks/sieve-vanilla/index.html
rename to tests/benchmarks/sieve-vanilla/index.html
diff --git a/benchmarks/sieve-vanilla/index.js b/tests/benchmarks/sieve-vanilla/index.js
similarity index 100%
rename from benchmarks/sieve-vanilla/index.js
rename to tests/benchmarks/sieve-vanilla/index.js
diff --git a/benchmarks/sieve-vanilla/sieve-vanilla.wat b/tests/benchmarks/sieve-vanilla/sieve-vanilla.wat
similarity index 99%
rename from benchmarks/sieve-vanilla/sieve-vanilla.wat
rename to tests/benchmarks/sieve-vanilla/sieve-vanilla.wat
index 42178ea..2268680 100644
--- a/benchmarks/sieve-vanilla/sieve-vanilla.wat
+++ b/tests/benchmarks/sieve-vanilla/sieve-vanilla.wat
@@ -1,6 +1,6 @@
 (module
   (import "js" "print" (func $print (param i32)))
-  (memory 4096)
+  (memory 8192)
   (func $sieve (export "sieve") (param $n i32) (result i32)
     (local $i i32)
     (local $j i32)