I think that it meets the guidelines. It is a matter of reposting it with the prefix "Show HN"?

Good question. The actor model communicate via mailboxes, which execute callbacks sort to say. Essentially this produces the callback hell.

Taking away the initializations, this is the ping pong example in Erlang:

https://gist.github.com/torgeir/3978785

    ping(Receiver) ->
      { pong, Receiver } ! { ping, self() },
      receive
        pong -> io:format("Got pong!~n")
    end.

    pong() ->
      receive

        finished ->
          io:format("Pong finished~n");

        { ping, Sender } ->
          io:format("Got ping!~n"),
          Sender ! pong,
          pong()
      end.

This is the same in Transient: A message is sent to the receiver print "ping" in his console, then send a message to the sender, that print "pong" in his console

    pingPong receiver=  
      wormhole receiver $ do
          teleport
          lliftIO $ print "ping"
          teleport
          lliftIO $ print "pong"

You see that the ping and the pong are composed "monadically" in a single piece of code. The flow is quite understandable.

Just start the program in both nodes (initialization code is not shown)

This is a distributed program but I can compose this program with any other.

This other program stream integers from a remote node from 1 in increasing order:

    streamFrom node= 
       wormhole  node $ do
                  teleport
                  r <- threads 1 $ choose[1..]
                  threadDelay 1000000
                  teleport
                  return r

I can compose both distributed programs:

    composed node =  do
         r <-  (pingPong node >> return 0) <|>  streamFrom node
         lliftIO $ case r of
              0 -> putStrLn "pong was received"
              n -> print n

`composed` print the numbers received and the "pong" messages followed by "pong was received" in the console of the caller.

Haxl is a great Haskell library and probably the biggest contribution to getting Haskell out of the academia. It is optimized for fetching querying and caching data sources, among other things, making use of parallelism and concurrency while maintaining a level close to the domain problem.

Haxl uses a controlling thread that spawn worker threads and is single purpose. No distributed computing in the strong sense.

transient has no controlling thread. It run multiple threads and is general purpose.

The alternative operator <|> can spawn parallel threads/or parallel distributed computations that follow the flow of the monad. for example:

    r <-   runAt node1 foo <|>  runAt node2 bar
    lliftIO $ print r

will print two results, since there are two thread generated. Well there may be any number of results because any of the two branches may generate multiple threads or none, since they can stop without any result.

(if we need a single thread, we can use `collect`)

note that `foo` and `bar` are distributed programs as well, so they can invoke other nodes and so on.