diff --git a/instapaper/__init__.py b/instapaper/__init__.py
new file mode 100644
index 0000000..b1c479b
--- /dev/null
+++ b/instapaper/__init__.py
@@ -0,0 +1,135 @@
+from datetime import datetime
+import json
+from pathlib import Path
+import pytz
+from typing import NamedTuple, Optional, List, Dict, Iterator, Tuple
+from collections import OrderedDict
+
+from kython import group_by_key
+
+BDIR = Path('/L/backups/instapaper/')
+
+Bid = str
+Hid = str
+
+class Highlight(NamedTuple):
+    dt: datetime
+    uid: Hid
+    bid: Bid
+    text: str
+    note: Optional[str]
+    url: str
+    title: str
+
+    @property
+    def instapaper_link(self) -> str:
+        return f'https://www.instapaper.com/read/{self.bid}/{self.uid}'
+
+
+class Bookmark(NamedTuple):
+    bid: Bid
+    dt: datetime
+    url: str
+    title: str
+
+    @property
+    def instapaper_link(self) -> str:
+        return f'https://www.instapaper.com/read/{self.bid}'
+
+
+class Page(NamedTuple):
+    bookmark: Bookmark
+    highlights: List[Highlight]
+
+def get_files():
+    return sorted(f for f in BDIR.iterdir() if f.suffix == '.json')
+
+def dkey(x):
+    return lambda d: d[x]
+
+
+def make_dt(time) -> datetime:
+    return pytz.utc.localize(datetime.utcfromtimestamp(time))
+
+
+BDict = Dict[Bid, Bookmark]
+HDict = Dict[Hid, Highlight]
+
+
+def get_stuff(limit=0) -> Tuple[BDict, HDict]:
+    all_bks: BDict = OrderedDict()
+    all_hls: HDict = OrderedDict()
+    # TODO can restore url by bookmark id
+    for f in get_files()[-limit:]:
+        with f.open('r') as fo:
+            j = json.load(fo)
+        for b in sorted(j['bookmarks'], key=dkey('time')):
+            bid = str(b['bookmark_id'])
+            prevb = all_bks.get(bid, None)
+            # assert prev is None or prev == b, '%s vs %s' % (prev, b)
+            # TODO shit, ok progress can change apparently
+            all_bks[bid] = Bookmark(
+                bid=bid,
+                dt=make_dt(b['time']),
+                url=b['url'],
+                title=b['title'],
+            )
+        hls = j['highlights']
+        for h in sorted(hls, key=dkey('time')):
+            hid = str(h['highlight_id'])
+            bid = str(h['bookmark_id'])
+            # TODO just reference to bookmark in hightlight?
+            bk = all_bks[bid]
+            h = Highlight(
+                uid=hid,
+                bid=bk.bid,
+                dt=make_dt(h['time']),
+                text=h['text'],
+                note=h['note'],
+                url=bk.url,
+                title=bk.title,
+            )
+            prev = all_hls.get(hid, None)
+            assert prev is None or prev == h
+            all_hls[hid] = h
+
+    return all_bks, all_hls
+
+def iter_highlights(**kwargs) -> Iterator[Highlight]:
+    return iter(get_stuff(**kwargs)[1].values())
+
+
+def get_highlights(**kwargs) -> List[Highlight]:
+    return list(iter_highlights(**kwargs))
+
+
+def get_todos() -> List[Highlight]:
+    def is_todo(h):
+        return h.note is not None and h.note.lstrip().lower().startswith('todo')
+    return list(filter(is_todo, iter_highlights()))
+
+
+def get_pages(**kwargs) -> List[Page]:
+    bms, hls = get_stuff(**kwargs)
+    groups = group_by_key(hls.values(), key=lambda h: h.bid)
+    pages = []
+    # TODO how to make sure there are no dangling bookmarks?
+    for bid, bm in bms.items():
+        pages.append(Page(
+            bookmark=bm,
+            highlights=sorted(groups.get(bid, []), key=lambda b: b.dt),
+        ))
+    return pages
+
+
+def test_get_todos():
+    for t in get_todos():
+        print(t)
+
+
+def main():
+    for h in get_todos():
+        print(h)
+
+if __name__ == '__main__':
+    main()