diff --git a/lib/coffee_script/lexer.js b/lib/coffee_script/lexer.js
new file mode 100644
index 00000000..9c01af22
--- /dev/null
+++ b/lib/coffee_script/lexer.js
@@ -0,0 +1,96 @@
+(function(){
+  var lex, sys;
+  sys = require('sys');
+  // The lexer reads a stream of CoffeeScript and divvys it up into tagged
+  // tokens. A minor bit of the ambiguity in the grammar has been avoided by
+  // pushing some extra smarts into the Lexer.
+  exports.Lexer = (lex = function lex() {  });
+  // The list of keywords passed verbatim to the parser.
+  lex.KEYWORDS = ["if", "else", "then", "unless", "true", "false", "yes", "no", "on", "off", "and", "or", "is", "isnt", "not", "new", "return", "arguments", "try", "catch", "finally", "throw", "break", "continue", "for", "in", "of", "by", "where", "while", "delete", "instanceof", "typeof", "switch", "when", "super", "extends"];
+  // Token matching regexes.
+  lex.IDENTIFIER = /^([a-zA-Z$_](\w|\$)*)/;
+  lex.NUMBER = /^(\b((0(x|X)[0-9a-fA-F]+)|([0-9]+(\.[0-9]+)?(e[+\-]?[0-9]+)?)))\b/i;
+  lex.STRING = /^(""|''|"(.*?)([^\\]|\\\\)"|'(.*?)([^\\]|\\\\)')/m;
+  lex.HEREDOC = /^("{6}|'{6}|"{3}\n?(.*?)\n?([ \t]*)"{3}|'{3}\n?(.*?)\n?([ \t]*)'{3})/m;
+  lex.JS = /^(``|`(.*?)([^\\]|\\\\)`)/m;
+  lex.OPERATOR = /^([+\*&|\/\-%=<>:!?]+)/;
+  lex.WHITESPACE = /^([ \t]+)/;
+  lex.COMMENT = /^(((\n?[ \t]*)?#.*$)+)/;
+  lex.CODE = /^((-|=)>)/;
+  lex.REGEX = /^(\/(.*?)([^\\]|\\\\)\/[imgy]{0,4})/;
+  lex.MULTI_DENT = /^((\n([ \t]*))+)(\.)?/;
+  lex.LAST_DENT = /\n([ \t]*)/;
+  lex.ASSIGNMENT = /^(:|=)$/;
+  // Token cleaning regexes.
+  lex.JS_CLEANER = /(^`|`$)/;
+  lex.MULTILINER = /\n/;
+  lex.STRING_NEWLINES = /\n[ \t]*/;
+  lex.COMMENT_CLEANER = /(^[ \t]*#|\n[ \t]*$)/;
+  lex.NO_NEWLINE = /^([+\*&|\/\-%=<>:!.\\][<>=&|]*|and|or|is|isnt|not|delete|typeof|instanceof)$/;
+  lex.HEREDOC_INDENT = /^[ \t]+/;
+  // Tokens which a regular expression will never immediately follow, but which
+  // a division operator might.
+  // See: http://www.mozilla.org/js/language/js20-2002-04/rationale/syntax.html#regular-expressions
+  lex.NOT_REGEX = ['IDENTIFIER', 'NUMBER', 'REGEX', 'STRING', ')', '++', '--', ']', '}', 'FALSE', 'NULL', 'TRUE'];
+  // Tokens which could legitimately be invoked or indexed.
+  lex.CALLABLE = ['IDENTIFIER', 'SUPER', ')', ']', '}', 'STRING'];
+  // Scan by attempting to match tokens one character at a time. Slow and steady.
+  lex.prototype.tokenize = function tokenize(code) {
+    this.code = code.chomp;
+    // Cleanup code by remove extra line breaks
+    this.i = 0;
+    // Current character position we're parsing
+    this.line = 1;
+    // The current line.
+    this.indent = 0;
+    // The current indent level.
+    this.indents = [];
+    // The stack of all indent levels we are currently within.
+    this.tokens = [];
+    // Collection of all parsed tokens in the form [:TOKEN_TYPE, value]
+    this.spaced = nil;
+    // The last value that has a space following it.
+    while (this.i < this.code.length) {
+      this.chunk = this.code.slice(this.i, -1 + 1);
+      this.extract_next_token();
+    }
+    if (process.ENV['VERBOSE']) {
+      sys.puts("original stream: #{@tokens.inspect}");
+    }
+    this.close_indentation();
+    return (new Rewriter()).rewrite(this.tokens);
+  };
+  // At every position, run through this list of attempted matches,
+  // short-circuiting if any of them succeed.
+  lex.prototype.extract_next_token = function extract_next_token() {
+    if (this.identifier_token()) {
+      return null;
+    }
+    if (this.number_token()) {
+      return null;
+    }
+    if (this.heredoc_token()) {
+      return null;
+    }
+    if (this.string_token()) {
+      return null;
+    }
+    if (this.js_token()) {
+      return null;
+    }
+    if (this.regex_token()) {
+      return null;
+    }
+    if (this.indent_token()) {
+      return null;
+    }
+    if (this.comment_token()) {
+      return null;
+    }
+    if (this.whitespace_token()) {
+      return null;
+    }
+    return this.literal_token();
+  };
+  // Tokenizers ==========================================================
+})();
\ No newline at end of file
diff --git a/src/lexer.coffee b/src/lexer.coffee
new file mode 100644
index 00000000..0bf041c1
--- /dev/null
+++ b/src/lexer.coffee
@@ -0,0 +1,114 @@
+sys: require 'sys'
+
+# The lexer reads a stream of CoffeeScript and divvys it up into tagged
+# tokens. A minor bit of the ambiguity in the grammar has been avoided by
+# pushing some extra smarts into the Lexer.
+exports.Lexer: lex: ->
+
+# The list of keywords passed verbatim to the parser.
+lex.KEYWORDS: [
+  "if", "else", "then", "unless",
+  "true", "false", "yes", "no", "on", "off",
+  "and", "or", "is", "isnt", "not",
+  "new", "return", "arguments",
+  "try", "catch", "finally", "throw",
+  "break", "continue",
+  "for", "in", "of", "by", "where", "while",
+  "delete", "instanceof", "typeof",
+  "switch", "when",
+  "super", "extends"
+]
+
+# Token matching regexes.
+lex.IDENTIFIER : /^([a-zA-Z$_](\w|\$)*)/
+lex.NUMBER     : /^(\b((0(x|X)[0-9a-fA-F]+)|([0-9]+(\.[0-9]+)?(e[+\-]?[0-9]+)?)))\b/i
+lex.STRING     : /^(""|''|"(.*?)([^\\]|\\\\)"|'(.*?)([^\\]|\\\\)')/m
+lex.HEREDOC    : /^("{6}|'{6}|"{3}\n?(.*?)\n?([ \t]*)"{3}|'{3}\n?(.*?)\n?([ \t]*)'{3})/m
+lex.JS         : /^(``|`(.*?)([^\\]|\\\\)`)/m
+lex.OPERATOR   : /^([+\*&|\/\-%=<>:!?]+)/
+lex.WHITESPACE : /^([ \t]+)/
+lex.COMMENT    : /^(((\n?[ \t]*)?#.*$)+)/
+lex.CODE       : /^((-|=)>)/
+lex.REGEX      : /^(\/(.*?)([^\\]|\\\\)\/[imgy]{0,4})/
+lex.MULTI_DENT : /^((\n([ \t]*))+)(\.)?/
+lex.LAST_DENT  : /\n([ \t]*)/
+lex.ASSIGNMENT : /^(:|=)$/
+
+# Token cleaning regexes.
+lex.JS_CLEANER      : /(^`|`$)/
+lex.MULTILINER      : /\n/
+lex.STRING_NEWLINES : /\n[ \t]*/
+lex.COMMENT_CLEANER : /(^[ \t]*#|\n[ \t]*$)/
+lex.NO_NEWLINE      : /^([+\*&|\/\-%=<>:!.\\][<>=&|]*|and|or|is|isnt|not|delete|typeof|instanceof)$/
+lex.HEREDOC_INDENT  : /^[ \t]+/
+
+# Tokens which a regular expression will never immediately follow, but which
+# a division operator might.
+# See: http://www.mozilla.org/js/language/js20-2002-04/rationale/syntax.html#regular-expressions
+lex.NOT_REGEX: [
+  'IDENTIFIER', 'NUMBER', 'REGEX', 'STRING',
+  ')', '++', '--', ']', '}',
+  'FALSE', 'NULL', 'TRUE'
+]
+
+# Tokens which could legitimately be invoked or indexed.
+lex.CALLABLE: ['IDENTIFIER', 'SUPER', ')', ']', '}', 'STRING']
+
+# Scan by attempting to match tokens one character at a time. Slow and steady.
+lex::tokenize: (code) ->
+  this.code    : code.chomp # Cleanup code by remove extra line breaks
+  this.i       : 0          # Current character position we're parsing
+  this.line    : 1          # The current line.
+  this.indent  : 0          # The current indent level.
+  this.indents : []         # The stack of all indent levels we are currently within.
+  this.tokens  : []         # Collection of all parsed tokens in the form [:TOKEN_TYPE, value]
+  this.spaced  : nil        # The last value that has a space following it.
+  while this.i < this.code.length
+    this.chunk: this.code[this.i..-1]
+    this.extract_next_token()
+  sys.puts "original stream: #{@tokens.inspect}" if process.ENV['VERBOSE']
+  this.close_indentation()
+  (new Rewriter()).rewrite(this.tokens)
+
+# At every position, run through this list of attempted matches,
+# short-circuiting if any of them succeed.
+lex::extract_next_token: ->
+  return if this.identifier_token()
+  return if this.number_token()
+  return if this.heredoc_token()
+  return if this.string_token()
+  return if this.js_token()
+  return if this.regex_token()
+  return if this.indent_token()
+  return if this.comment_token()
+  return if this.whitespace_token()
+  return    this.literal_token()
+
+# Tokenizers ==========================================================
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+