diff --git a/README.md b/README.md
new file mode 100644
index 0000000..9427ea4
--- /dev/null
+++ b/README.md
@@ -0,0 +1,68 @@
+videostreamer provides a way to stream video from an input source to HTTP. It
+remuxes a given video input into an MP4 container which it then streams
+endlessly to any connecting clients. This provides the ability to stream an
+input source that may have limited connections or not accessible via HTTP.
+
+
+# Background
+A friend has a camera that publishes an RTSP feed containing h264 video. It
+permits a limited number of connections, and being RTSP is not easily
+accessible on the web, such as in a regular HTML5 `<video>` element. This
+project solves both of these problems. It reads the RTSP feed as input, remuxes
+the h264 video into an MP4 container, and streams that to HTTP clients. No
+matter how many clients are streaming the video, videostreamer only opens one
+RTSP connection.
+
+Happily, browsers support h264 inside an MP4 container, so no transcoding is
+necessary. However, in order to stream the MP4, it is fragmented using
+libavformat's `frag_keyframe` option. I found that in Firefox, I also had to
+set the `empty_moov` option.
+
+
+# Difference from audiostreamer
+I have a similar project for streaming audio called
+[audiostreamer](https://github.com/horgh/audiostreamer). I wrote it before this
+one, and seeing it was succesful and gave me experience with the ffmpeg
+libraries, I decided to build this too.
+
+Initially I was going to make one larger project to stream either video or audio
+(or both), but I found that video behaved a bit differently than audio.
+Specifically, audiostreamer is able to create a single MP3 container and then
+stream that to clients, starting at any frame position. For the video container
+I wanted to use, I found this was not possible. First, I always needed to send
+a header to each client, which meant I needed to keep the header around in some
+way to send to every client. I found that I could not gain access to the header
+from the libavformat API. Second, I was not able to read individual frames out
+and send them independently like I could do with MP3. This meant it was
+impossible to know where in the MP4 container's output I could begin sending to
+a client that connected when another client was already streaming.
+
+Since it became clear that I would have to structure the program quite
+differently, and I was happy with how audiostreamer was, I decided to split them
+into two projects.
+
+
+# Requirements
+  * ffmpeg (libavcodec, libavformat, libavdevice, libavutil, libswresample). I
+    developed using 3.2.2. On Debian this is in the package libavutil-dev.
+  * C compiler. I developed using gcc 6.2.1.
+  * Go. I developed using 1.7.3.
+
+
+# Installation
+  * go get github.com/horgh/videostreamer
+  * go build
+  * Place index.html somewhere accessible. Update the `<video>` element src
+    attribute.
+  * Run the daemon. Its usage output shows the possible flags. There is no
+    configuration file.
+
+
+# Components
+  * videostreamer: A daemon. The main component of the project.
+  * index.html: A small sample website with a `<video>` element which
+    demonstrates how to stream from the daemon.
+  * videostreamer.h: A library using the ffmpeg libraries to read a video input
+    and remux and write to another format.
+  * cmd/remux_example: A C program demonstrating using videostreamer.h. It
+    remuxes from a given RTSP input to an MP4 file.