diff --git a/downloader/downloader/twitch.py b/downloader/downloader/twitch.py
new file mode 100644
index 0000000..fc0d425
--- /dev/null
+++ b/downloader/downloader/twitch.py
@@ -0,0 +1,104 @@
+
+import logging
+import random
+
+import requests
+
+import hls_playlist
+
+
+def get_master_playlist(channel):
+	"""Get the master playlist for given channel from twitch"""
+	resp = requests.get(
+		"https://api.twitch.tv/api/channels/{}/access_token.json".format(channel),
+		params={'as3': 't'},
+		headers={
+			'Accept': 'application/vnd.twitchtv.v3+json',
+			'Client-ID': 'pwkzresl8kj2rdj6g7bvxl9ys1wly3j',
+		},
+	)
+	resp.raise_for_status() # getting access token
+	token = resp.json()
+	resp = requests.get(
+		"https://usher.ttvnw.net/api/channel/hls/{}.m3u8".format(channel),
+		params={
+			# Taken from streamlink. Unsure what's needed and what changing things can do.
+			"player": "twitchweb",
+			"p": random.randrange(1000000),
+			"type": "any",
+			"allow_source": "true",
+			"allow_audio_only": "true",
+			"allow_spectre": "false",
+			"fast_bread": "True",
+			"sig": token["sig"],
+			"token": token["token"],
+			# Also observed in the wild but not used in streamlink:
+			# "playlist_include_framerate": "true"
+			# "reassignments_supported": "true"
+		},
+	)
+	resp.raise_for_status() # getting master playlist
+	playlist = hls_playlist.load(resp.text, base_uri=resp.url)
+	return playlist
+
+
+def get_media_playlist_uris(master_playlist, target_qualities):
+	"""From a master playlist, extract URIs of media playlists of interest.
+	Returns {stream name: uri}.
+	Note this is not a general method for all HLS streams, and makes twitch-specific assumptions,
+	though we try to check and emit warnings if these assumptions are broken.
+	"""
+	# Twitch master playlists are observed to have the following form:
+	#   The first listed variant is the source playlist and has "(source)" in the name.
+	#   Other variants are listed in order of quality from highest to lowest, followed by audio_only.
+	#   These transcoded variants are named "Hp[R]" where H is the vertical resolution and
+	#   optionally R is the frame rate. R is elided if == 30. Examples: 720p60, 720p, 480p, 360p, 160p
+	#   These variants are observed to only ever have one rendition, type video, which contains the name
+	#   but no URI. The URI in the main variant entry is the one to use. This is true even of the
+	#   "audio_only" stream.
+	#   Streams without transcoding options only show source and audio_only.
+	# We return the source stream in addition to any in target_qualities that is found.
+
+	def variant_name(variant):
+		names = set(media.name for media in variant.media if media.type == "VIDEO" and media.name)
+		if not names:
+			logging.warning("Variant {} has no named video renditions, can't determine name".format(variant))
+			return None
+		if len(names) > 1:
+			logging.warning("Variant {} has multiple possible names, picking one arbitrarily".format(variant))
+		return list(names)[0]
+
+	if not master_playlist.playlists:
+		raise ValueError("Master playlist has no variants")
+
+	for variant in master_playlist.playlists:
+		if any(media.uri for media in variant.medias):
+			logging.warning("Variant has a rendition with its own URI: {}".format(variant))
+
+	by_name = {variant_name(variant): variant for variant in master_playlist.playlists}
+
+	source_candidates = [name for name in by_name.keys() if "(source)" in name]
+	if len(source_candidates) != 1:
+		raise ValueError("Can't find source stream, not exactly one candidate. Candidates: {}, playlist: {}".format(
+			source_candidates, master_playlist,
+		))
+	source = by_name[source_candidates[0]]
+
+	variants = {name: variant for name, variant in by_name.items() if name in target_qualities}
+	variants["source"] = source
+
+	return {name: variant.uri for name, variant in variants.items()}
+
+
+def get_media_playlist(uri):
+	resp = requests.get(uri)
+	resp.raise_for_status()
+	return hls_playlist.load(resp.text, base_uri=resp.url)
+
+
+def get_segment(uri):
+	# Future work: Make this stream directly to a file + hash it as we go,
+	# to avoid keeping it all in memory at once.
+	resp = requests.get(uri)
+	resp.raise_for_status()
+	return resp.content