diff --git a/examples/ai_chat.ml b/examples/ai_chat.ml
new file mode 100644
index 0000000..ff792d9
--- /dev/null
+++ b/examples/ai_chat.ml
@@ -0,0 +1,106 @@
+class read_line ~term ~history ~state =
+  object (self)
+    inherit LTerm_read_line.read_line ~history ()
+
+    inherit [Zed_string.t] LTerm_read_line.term term
+
+    method! show_box = false
+
+    initializer
+    self#set_prompt
+      (React.S.const (LTerm_text.of_string @@ Zed_string.of_utf8 "\\>"))
+  end
+
+let rec loop term history state =
+  let%lwt input =
+    Lwt.catch
+      (fun () ->
+        let rl =
+          new read_line ~term ~history:(LTerm_history.contents history) ~state
+        in
+        let%lwt command = rl#run in
+        Lwt.return_some command )
+      (function Sys.Break -> Lwt.return None | exn -> Lwt.fail exn)
+  in
+  match input with
+  | Some input ->
+      let message =
+        Openai.Chat_completion_request_message.make
+          ~role:`User
+          ~content:(Zed_string.to_utf8 input)
+          ()
+      in
+      let state = state @ [message] in
+      let req =
+        Openai.Create_chat_completion_request.make
+          ~model:"gpt-3.5-turbo"
+          ~messages:state
+          ()
+      in
+      (* n = 1 so we can assume the list is always shaped like this *)
+      let%lwt[@warning "-8"] {choices= [{message= message'; _}]; _} =
+        Openai.Open_ai_api.create_chat_completion
+          ~create_chat_completion_request_t:
+            {req with messages= state; n= Some 1l}
+      in
+      let state =
+        state @ [{role= message'.role; content= message'.content; name= None}]
+      in
+      let label = new LTerm_widget.label message'.content in
+      let () = label#set_alignment LTerm_geom.H_align_right in
+      let%lwt () =
+        let term_width = LTerm.size term |> LTerm_geom.cols in
+        let text_width = if term_width > 80 then 80 else term_width in
+        let message'' =
+          let lines = String.split_on_char '\n' message'.content in
+          let rec aux = function
+            | [] -> []
+            | line :: lines ->
+                if String.length line > text_width then
+                  let line' = String.sub line 0 text_width in
+                  let line'' = String.sub line text_width (String.length line - text_width) in
+                  line' :: aux (line'' :: lines)
+                else line :: aux lines
+          in
+          aux lines |> String.concat "\n"
+        in
+        LTerm_text.(
+          eval [
+            B_bold true;
+            B_fg LTerm_style.black;
+            B_bg (LTerm_style.rgb 128 255 128);
+            S message'';
+            E_bg;
+            E_fg;
+            E_bold;
+          ])
+        |> LTerm.fprintls term
+      in
+      LTerm_history.add history input ;
+      loop term history state
+  | None ->
+      loop term history state
+
+let chatgpt initial_state =
+  let%lwt () = LTerm_inputrc.load () in
+  Lwt.catch
+    (fun () ->
+      let state = initial_state in
+      let%lwt term = Lazy.force LTerm.stdout in
+      loop term (LTerm_history.create []) state )
+    (function LTerm_read_line.Interrupt -> Lwt.return () | exn -> Lwt.fail exn)
+
+let _ =
+  let system_prompt =
+    Openai.Chat_completion_request_message.make
+      ~role:`System
+      ~content:"You are a general-purpose assistant who provides detailed, helpful responses."
+      ()
+  in
+  let init_message =
+    Openai.Chat_completion_request_message.make
+      ~role:`User
+      ~content:"Hello! What are you?"
+      ()
+  in
+  Lwt_main.run (chatgpt [system_prompt; init_message])
diff --git a/examples/dune b/examples/dune
new file mode 100644
index 0000000..d8e9689
--- /dev/null
+++ b/examples/dune
@@ -0,0 +1,14 @@
+(executable
+ (public_name ai-chat)
+ (name ai_chat)
+ (libraries openai lwt lwt.unix lambda-term react)
+ (flags -g)
+ (preprocess
+  (pps
+   ppx_deriving_yojson
+   ppx_deriving.show
+   ppx_deriving.eq
+   ppx_deriving.ord
+   ppx_deriving.std
+   ez_api.ppx
+   lwt_ppx)))