An address looks like this: BM-orkCbppXWSqPpAxnz6jnfTZ2djb5pJKDb (this one belongs to an echo server). People who got used to Base58-encoded Bitcoin addresses will find this familiar.
Bitmessage does not do any OTR tricks (I myself do not think OTR is very important), but guarantees anonymous routing probably even better than Tor: every message is routed to everyone randomly without any particular route whatsoever. In addition, protocol can be extended to wrap messages with some random person's key, so you get double-clouded routing.
So, Bitmessage might give you even better tangible privacy over Tor, while not giving any of disputable value of OTR.
PS. Why I think OTR is of little value.
OTR is a trick Alice uses to release signing keys after talking to Bob. Bob can retain the info on his disk and be interrogated, but Alice sort of "proves" that anyone could produce that data, so she's innocent. In other words, OTR is not about hiding any information, but about telling the court that "i was smart, you can't prove it was written by me". Logically OTR makes sense, but without being tested in court (which is not a department of logic) it's worthless. And I believe, when it comes to a trial, your adversaries will have plenty of parallel construction in addition to judge's "common sense" to invalidate any trickery that your might have employed. In other words: if you talked to Bob and his data is taken by NSA, you are doomed, with or without OTR.
EDIT: I meant "deniability" part of OTR. "Perfect forward secrecy" is indeed valuable, but not implemented yet.
More info: https://bitmessage.org
The idea behind Bitmessage is that messages are fully encrypted (including recipient's address) with per-message random DH key derived from recipient's key. Then message is transmitted to everyone in a p2p network. Every node tries to decrypt the message with its key and if it succeeds - that's the message for them. If not, it passes the message to other nodes. Propagation time is quick, but to prevent DoS, each message goes with an expensive one-time proof-of-work proportional to the message size. Another measure is artificial separation of messages and nodes in "streams" so that message is only propagated within a smaller part of the network. As network grows this will not hurt privacy, but will keep amount of data flowing around in check. Bitmessage in principle is like email (that is checked every few minutes, not seconds) rather than real-time chat. But the idea could be brought to real-time chats too if we solve DoS and bandwidth issues in a different way.