diff --git a/openai/init.moon b/openai/init.moon
index 072242d..60fc3e4 100644
--- a/openai/init.moon
+++ b/openai/init.moon
@@ -82,19 +82,6 @@ parse_chat_response = types.partial {
 -- }
 
 
-parse_completion_chunk = types.partial {
-  object: "chat.completion.chunk"
-  -- not sure of the whole range of chunks, so for now we strictly parse an append
-  choices: types.shape {
-    types.partial {
-      delta: types.shape {
-        "content": types.string\tag "content"
-      }
-      index: types.number\tag "index"
-    }
-  }
-}
-
 -- lpeg pattern to read a json data block from the front of a string, returns
 -- the json blob and the rest of the string if it could parse one
 consume_json_head = do
@@ -183,20 +170,8 @@ class ChatSession
 
       return nil, err_msg, response
 
-    -- if we are streaming we need to pase the entire fragmented response
     if stream_callback
-      assert type(response) == "string",
-        "Expected string response from streaming output"
-
-      parts = {}
-      f = @client\create_stream_filter (c) ->
-        table.insert parts, c.content
-
-      f response
-      message = {
-        role: "assistant"
-        content: table.concat parts
-      }
+      message = response.choices[1].message
 
       if append_response
         @append_message message
@@ -250,7 +225,7 @@ class OpenAI
             break
 
           accumulation_buffer = rest
-          if chunk = parse_completion_chunk cjson.decode json_blob
+          if chunk = cjson.decode json_blob
             chunk_callback chunk
 
       ...
@@ -273,10 +248,7 @@ class OpenAI
       for k,v in pairs opts
         payload[k] = v
 
-    stream_filter = if payload.stream
-      @create_stream_filter chunk_callback
-
-    @_request "POST", "/chat/completions", payload, nil, stream_filter
+    @_request "POST", "/chat/completions", payload, nil, if payload.stream then chunk_callback else nil
 
   -- call /completions
   -- opts: additional parameters as described in https://platform.openai.com/docs/api-reference/completions
@@ -362,7 +334,7 @@ class OpenAI
   image_generation: (params) =>
     @_request "POST", "/images/generations", params
 
-  _request: (method, path, payload, more_headers, stream_fn) =>
+  _request: (method, path, payload, more_headers, chunk_callback) =>
     assert path, "missing path"
     assert method, "missing method"
 
@@ -392,7 +364,17 @@ class OpenAI
 
     sink = ltn12.sink.table out
 
-    if stream_fn
+    parts = {}
+    if chunk_callback
+      stream_fn = @create_stream_filter (c) ->
+        c0 = c.choices[1]
+        part = parts[c0.index] or {}
+        part.data = c
+        part.finish_reason = c0.finish_reason
+        parts[c0.index] = part
+        if c0.delta.content and c0.delta.content ~= cjson.null
+          table.insert part, c0.delta.content
+        chunk_callback(c)
       sink = ltn12.sink.chain stream_fn, sink
 
     _, status, out_headers = @get_http!.request {
@@ -403,6 +385,23 @@ class OpenAI
       :headers
     }
 
+    if status == 200 and chunk_callback
+      choices = {}
+      index = 0
+      local data
+      while parts[index]
+        part = parts[index]
+        data = part.data
+        message = {
+          role: "assistant"
+          content: table.concat part
+        }
+        choices[index+1] = { :index, :message, finish_reason: part.finish_reason }
+        index += 1
+      data.object = "chat.completion"
+      data.choices = choices
+      return status, data, out_headers
+
     response = table.concat out
     pcall -> response = cjson.decode response
     status, response, out_headers