diff --git a/Llama.hs b/Llama.hs
--- a/Llama.hs
+++ b/Llama.hs
@@ -1,18 +1,54 @@
-{-# LANGUAGE OverloadedStrings #-}
+{-# LANGUAGE OverloadedStrings, DeriveGeneric #-}
 
 module Llama where
 
-import Network.HTTP.Conduit
+import Control.Monad.IO.Class (liftIO)
 import Data.Aeson
 import Data.Text (Text)
-import Control.Monad.IO.Class (liftIO)
+import GHC.Generics
+import Network.HTTP.Conduit
 import System.IO (hPutStrLn, stderr)
 
+data Role = System | User | CustomRole Text deriving Show
+instance ToJSON Role where
+  toJSON System = "system"
+  toJSON User = "system"
+  toJSON (CustomRole t) = String t
+
+data LlamaMessage = LlamaMessage
+  { role :: Role
+  , content :: Text
+  } deriving (Show)
+
+instance ToJSON LlamaMessage where
+  toJSON m =
+    object
+      [ "role" .= role m
+      , "content" .= content m
+      ]
+
+data LlamaApplyTemplateRequest = LlamaApplyTemplateRequest
+  { messages :: [LlamaMessage]
+  } deriving (Show, Generic)
+instance ToJSON LlamaApplyTemplateRequest
+
+--data LlamaApplyTemplateResponse = LlamaApplyTemplateResponse
+--  { prompt :: Text
+--  } deriving (Show)
+
 -- Llama request and response
 data LlamaRequest = LlamaRequest
   { prompt :: Text
   } deriving (Show)
 
+instance FromJSON LlamaRequest where
+  parseJSON = withObject "LlamaRequest" $ \v -> LlamaRequest
+    <$> v .: "prompt"
+
+instance ToJSON LlamaRequest where
+  toJSON (LlamaRequest p) =
+    object ["prompt" .= p]
+
 data LlamaResponse = LlamaResponse
   { generatedText :: Text
   } deriving (Show)
@@ -21,9 +57,20 @@
   parseJSON = withObject "LlamaResponse" $ \v -> LlamaResponse
     <$> v .: "content"
 
-instance ToJSON LlamaRequest where
-  toJSON (LlamaRequest p) =
-    object ["prompt" .= p]
+applyTemplate :: Manager -> LlamaApplyTemplateRequest -> IO (Maybe Text)
+applyTemplate manager input = do
+  let request = parseRequest_ "http://localhost:8080/apply-template"
+      body = encode input
+      req = request { method = "POST"
+                    , requestBody = RequestBodyLBS body
+                    , requestHeaders = [("Content-Type", "application/json")]
+                    }
+  response <- httpLbs req manager
+  case decode (responseBody response) of
+    Just (LlamaRequest text) -> return (Just text)
+    Nothing -> do
+      liftIO $ hPutStrLn stderr "Failed to decode Llama response"
+      return Nothing
 
 -- Function to send a message to the Llama model
 sendToLlama :: Manager -> Text -> IO (Maybe Text)
@@ -45,3 +92,11 @@
 llama input = do
   manager <- liftIO $ newManager tlsManagerSettings { managerResponseTimeout = responseTimeoutNone }
   sendToLlama manager input
+
+llamaTemplated :: LlamaApplyTemplateRequest -> IO (Maybe Text)
+llamaTemplated input = do
+  manager <- liftIO $ newManager tlsManagerSettings { managerResponseTimeout = responseTimeoutNone }
+  res <- applyTemplate manager input
+  case res of
+    Just text -> sendToLlama manager text
+    _ -> pure Nothing
diff --git a/llama-cpp-haskell.cabal b/llama-cpp-haskell.cabal
--- a/llama-cpp-haskell.cabal
+++ b/llama-cpp-haskell.cabal
@@ -1,6 +1,6 @@
 cabal-version:      2.2
 name:               llama-cpp-haskell
-version:            0
+version:            0.0.1
 synopsis:           Haskell bindings for the llama.cpp llama-server
 -- description:
 license:            AGPL-3.0-only
@@ -19,7 +19,7 @@
 Source-repository this
   type:              git
   location:          https://github.com/l29ah/llama-cpp-haskell.git
-  tag:               0
+  tag:               0.0.1
 
 common stuff
     ghc-options: -Wall