diff --git a/config.sample.yaml b/config.sample.yaml
index 0bb8137..afeb7d0 100644
--- a/config.sample.yaml
+++ b/config.sample.yaml
@@ -39,6 +39,9 @@ reply_in_thread: true
 # For example, you can instruct the assistant with "You are a helpful assistant."
 #system_prompt:
 
+# Insert the system prompt before the most recent user input. Useful for threaded chats.
+force_system_prompt: false
+
 # Log the full response (prompt + response) at debug level.
 log_full_response: false
 
diff --git a/main.py b/main.py
index de07359..c17f217 100755
--- a/main.py
+++ b/main.py
@@ -54,7 +54,6 @@ check_config_value_exists(config_data, 'openai_api_key')
 check_config_value_exists(config_data, 'openai_model')
 check_config_value_exists(config_data, 'data_storage')
 
-
 # check_config_value_exists(config_data, 'autojoin_rooms')
 
 def retry(msg=None):
@@ -89,7 +88,18 @@ async def main():
     storage = Storage(Path(config_data['data_storage'], 'matrixgpt.db'))
 
     # Set up event callbacks
-    callbacks = Callbacks(client, storage, config_data['command_prefix'], openai_config, config_data.get('reply_in_thread', False), config_data['allowed_to_invite'], config_data['allowed_to_chat'], config_data.get('system_prompt'), log_full_response=config_data.get('log_full_response', False))
+    callbacks = Callbacks(
+        client,
+        storage,
+        config_data['command_prefix'],
+        openai_config,
+        config_data.get('reply_in_thread', False),
+        config_data['allowed_to_invite'],
+        config_data['allowed_to_chat'],
+        config_data.get('system_prompt'),
+        log_full_response=config_data.get('log_full_response', False),
+        force_system_prompt=config_data.get('force_system_prompt', False)
+    )
     client.add_event_callback(callbacks.message, RoomMessageText)
     client.add_event_callback(callbacks.invite_event_filtered_callback, InviteMemberEvent)
     client.add_event_callback(callbacks.decryption_failure, MegolmEvent)
diff --git a/matrix_gpt/bot/callbacks.py b/matrix_gpt/bot/callbacks.py
index b528dcc..230aa41 100644
--- a/matrix_gpt/bot/callbacks.py
+++ b/matrix_gpt/bot/callbacks.py
@@ -13,7 +13,7 @@ logger = logging.getLogger('MatrixGPT')
 
 
 class Callbacks:
-    def __init__(self, client: AsyncClient, store: Storage, command_prefix: str, openai, reply_in_thread, allowed_to_invite, allowed_to_chat='all', system_prompt: str = None, log_full_response: bool = False):
+    def __init__(self, client: AsyncClient, store: Storage, command_prefix: str, openai, reply_in_thread, allowed_to_invite, allowed_to_chat='all', system_prompt: str = None, log_full_response: bool = False, force_system_prompt: bool = False):
         """
         Args:
             client: nio client used to interact with matrix.
@@ -33,6 +33,7 @@ class Callbacks:
         self.allowed_to_chat = allowed_to_chat
         self.system_prompt = system_prompt
         self.log_full_response = log_full_response
+        self.force_system_prompt = force_system_prompt
 
     async def message(self, room: MatrixRoom, event: RoomMessageText) -> None:
         """Callback for when a message event is received
@@ -91,7 +92,7 @@ class Callbacks:
             # message = Message(self.client, self.store, msg, room, event, self.reply_in_thread)
             # await message.process()
             api_data.append({'role': 'user', 'content': event.body})
-            await process_chat(self.client, room, event, api_data, self.store, self.openai, thread_root_id=thread_content[0].event_id, system_prompt=self.system_prompt, log_full_response=self.log_full_response)
+            await process_chat(self.client, room, event, api_data, self.store, self.openai, thread_root_id=thread_content[0].event_id, system_prompt=self.system_prompt, log_full_response=self.log_full_response, force_system_prompt=self.force_system_prompt)
             return
         elif msg.startswith(f'{self.command_prefix} ') or room.member_count == 2:
             # Otherwise if this is in a 1-1 with the bot or features a command prefix, treat it as a command.
diff --git a/matrix_gpt/bot/chat_functions.py b/matrix_gpt/bot/chat_functions.py
index 16cfb7c..5933e92 100644
--- a/matrix_gpt/bot/chat_functions.py
+++ b/matrix_gpt/bot/chat_functions.py
@@ -184,7 +184,7 @@ async def get_thread_content(client: AsyncClient, room: MatrixRoom, base_event:
     return messages
 
 
-async def process_chat(client, room, event, command, store, openai, thread_root_id: str = None, system_prompt: str = None, log_full_response: bool = False):
+async def process_chat(client, room, event, command, store, openai, thread_root_id: str = None, system_prompt: str = None, log_full_response: bool = False, force_system_prompt: bool = False):
     if not store.check_seen_event(event.event_id):
         await client.room_typing(room.room_id, typing_state=True, timeout=3000)
         # if self.reply_in_thread:
@@ -197,7 +197,14 @@ async def process_chat(client, room, event, command, store, openai, thread_root_
                 {'role': 'user', 'content': command},
             ]
         if system_prompt:
-            messages.insert(0, {"role": "system", "content": system_prompt}, )
+            messages.insert(0, {"role": "system", "content": system_prompt})
+        if force_system_prompt:
+            if messages[-1]['role'] == 'system':
+                messages[-1] = {"role": "system", "content": system_prompt}
+            else:
+                messages.insert(-1, {"role": "system", "content": system_prompt})
+
+        logger.info(messages)
 
         response = openai['openai'].ChatCompletion.create(
             model=openai['model'],