diff --git a/docs/bipack.md b/docs/bipack.md
new file mode 100644
index 0000000..1036cb1
--- /dev/null
+++ b/docs/bipack.md
@@ -0,0 +1,89 @@
+# Bipack: compact binary serialization
+
+## Why?
+
+Bipack was designed with the following main goals:
+
+### Be as compact as possible
+
+For this reason it is a binary notation, it uses binary form for decimal numbers and can use variery of encoding for
+integers:
+
+#### Varint
+
+Variable-length compact encoding is used internally in some cases. It uses a 0x80 bit in every byte to mark coninuation.
+See `object Varint`.
+
+#### Smartint
+
+Variable-length compact encoding for signed and unsigned integers use as few bytes as possible to encode integers. It is
+used automatically when serializing integers. It is slightly more sophisticated than straight `Varint`.
+
+### Do not reveal information about stored data
+
+Many extendable formats, like JSON, BSON, BOSS and may others are keeping data in key-value pairs. While it is good in
+many aspets, it has a clear disadvantages: it uses more space and it reveals inner data structure to the world. It is
+possible to unpack such formats with zero information about inner structure.
+
+Bipack does not store field names, so it is not possible to unpack or interpret it without knowledge of the data
+structure. Only probablistic analysis. Let's not make life of attacker easier :)
+
+### Allow upgrading data structures with backward compatibility
+
+The dark side of serialization formats of this kind is that you can't change the structures without either loosing
+backward compatibility with already serialzied data or using volumous boilerplate code to implement some sort of
+versioning.
+
+Not to waste space and reveal more information that needed Bipack allows extending classes marked as [@Extendable] to be
+extended with more data _appended to the end of list of fields with required defaul values_. For such classes Bipack
+stores number of actually serialized fields and atuomatically uses default values for non-serialized ones when unpacking
+old data.
+
+### Protect data with framing and CRC
+
+When needed, serialization lobrary allow to store/check CRC32 tag of the structure name with `@Framed` (can be overriden
+as usual with `@SerialName`), or be followed with CRC32 of the serialized binary data, that will be checked on
+deserialization, using `@CrcProtected`. This allows to check the data consistency out of the box and only where needed.
+
+# Usage
+
+Use kotlinx serializatino as usual. There are following Bipack-specific annotation at your service. All class annotations could be combined.
+
+## @Extendable
+
+Classes marked this way store number of fields. It allows to add to the class data more fields, to the end of list, with
+default initializers, keeping backward compatibility. For example if you have serialized:
+
+```kotlin
+@Serializable
+@Extendable
+data class foo(i: Int)
+```
+
+and then decided to add a field:
+
+```kotlin
+@Serializable
+@Extendable
+data class foo(val i: Int, bar: String = "buzz")
+```
+
+It adds 1 or more bytes to the serialized data (field counts in `Varint` format)
+
+Bipack will properly deserialize the data serialzied for an old version.
+
+## @CrcProtected
+
+Bipack will calculate and store CRC32 of serialized data at the end, and automatically check it on deserializing throwing `InvalidFrameCRCException` if it does not match.
+
+It adds 4 bytes to the serialized data.
+
+## @Framed
+
+Put the CRC32 of the serializing class name (`@SerialName` allows to change it as usual) and checks it on deserializing. Throws `InvalidFrameHeaderException` if it does not match.
+
+It adds 4 bytes to the serialized data.
+
+## @Unisgned
+
+This __field annontation__ allows to store __integer fields__ of any size more compact by not saving the sign. Could be applyed to both signed and unsigned integers of any size.