diff --git a/doc/proposal-2015-12-15.md b/doc/proposal-2015-12-15.md
new file mode 100644
index 0000000..e114588
--- /dev/null
+++ b/doc/proposal-2015-12-15.md
@@ -0,0 +1,55 @@
+Proposal proposed the 15th December of 2015
+===========================================
+We decided it would make sense to split the API into multiple levels, each
+getting more and more abstract, but targeting progressively more specific parts
+of the ricochet spec.
+
+### First Level: Version Negotiation
+The lowest level of the API will be the ricochet version negotiation. This is
+the only part of the spec that definetly won't change in future versions and can
+therefore doesn't target any specific version of the ricochet spec.
+
+The most abstract part of the API on this level will probably look something
+like this:
+```haskell
+runRicochet :: RSAKeyPair -> Map Word8 (Socket -> IO ()) -> IO ()
+```
+
+### Second Level: Packet Parsing
+The second level will implement the parsing of generic ricochet packets from
+`ByteString`s. This was specified in protocol version 1 and is therefore not
+version agnostic, but also seems like something that will stay the same in
+future versions.
+
+It will expose something like this:
+```haskell
+createPacketChan :: Socket -> IO (Chan Packet Packet)
+selectChannel :: Word16 -> Prism' Packet Packet
+```
+
+### Third Level: Ricochet Primitives
+This level will implement the primitives introduced by ricochet version 1,
+namely the authentication process for `im.ricochet.auth.hidden-service` and the
+protobuf parsing of the specified structures.
+
+### Fourth Level: High Level API
+This level will be what most users of this library will interact with. It's an
+high level interface to the ricochet protocol. The user will get a `Chan`
+carrying the `Packet` type for every connected peer. They will be able to apply
+prisms to it and filter / parse specific events that way.
+
+### Additional Feature: Custom Chan Type
+As seen throughout this document, we provide our own variant of
+Control.Concurrent.Chan. It will allow filtering, mapping and merging with
+ease, which is needed for the fourth API level. They'll be implemented like
+this:
+```haskell
+-- s is the source type
+data Chan s a = MkChan (Channel s) (s -> Maybe a)
+
+instance Functor (Chan s)
+
+transform :: (a -> Maybe b) -> Chan s a -> Chan s b
+-- The source Channel has to be identical for this to work properly
+merge :: Chan s a -> Chan s b -> Chan s (Either a b)
+```