diff --git a/sep/sep-004.rst b/sep/sep-004.rst
new file mode 100644
index 000000000..69edfa136
--- /dev/null
+++ b/sep/sep-004.rst
@@ -0,0 +1,85 @@
+=======  ====================================
+SEP      4
+Title    Library-Like API for quick scraping
+Author   Pablo Hoffman
+Created  2009-07-21
+Status   Archived
+=======  ====================================
+
+====================
+SEP-004: Library API
+====================
+.. note:: the library API has been implemented, but slightly different from
+          proposed in this SEP. You can run a Scrapy crawler inside a Twisted
+          reactor, but not outside it. 
+
+Introduction
+============
+
+It would be desirable for Scrapy to provide a quick, "light-weight" mechanism
+for implementing crawlers by just using callback functions. That way you could
+use Scrapy as any standard library (like you would use os.walk) in a script
+without the overhead of having to create an entire project from scratch.
+
+Proposed API
+============
+
+Here's a simple proof-of-concept code of such script:
+
+::
+
+   #!/usr/bin/env python
+   from scrapy.http import Request
+   from scrapy import Crawler
+
+   # a container to hold scraped items
+   scraped_items = []
+
+   def parse_start_page(response):
+       # collect urls to follow into urls_to_follow list
+       requests = [Request(url, callback=parse_other_page) for url in urls_to_follow]
+       return requests
+
+   def parse_other_page(response):
+       # ... parse items from response content ...
+       scraped_items.extend(parsed_items)
+
+   start_urls = ["http://www.example.com/start_page.html"]
+
+   cr = Crawler(start_urls, callback=parse_start_page)
+   cr.run() # blocking call - this populates scraped_items
+
+   print "%d items scraped" % len(scraped_items)
+   # ... do something more interesting with scraped_items ...
+
+The behaviour of the Scrapy crawler would be controller by the Scrapy settings,
+naturally, just like any typical scrapy project. But the default settings
+should be sufficient so as to not require adding any specific setting. But, at
+the same time, you could do it if you need to, say, for specifying a custom
+middleware.
+
+It shouldn't be hard to implement this API as all this functionality is a
+(small) subset of the current Scrapy functionality. At the same time, it would
+provide an additional incentive for newcomers.
+
+Crawler class
+=============
+
+The Crawler class would have the following instance arguments (most of them
+have been singletons so far):
+
+- engine
+- settings
+- spiders
+- extensions
+
+Spider Manager
+==============
+
+The role of the spider manager will be to "resolve" spiders from URLs and
+domains. Also, it should be moved outside scrapy.spider (and only BaseSpider
+left there).
+
+There is also the ``close_spider()`` method which is called for all closed
+spiders, even when they weren't resolved first by the spider manager. We need
+to decide what to do with this method.