How do i add system instructions to GoogleAIGeminiGenerator or GoogleAIGeminiChatGenerator

[JanB1989]

I simply do not understand how this is done. Docs say with vertexai you can set system instruction at generator instantiation. But this does not work for GoogleAIGeminiGenerator or GoogleAIGeminiChatGenerator.
Adding the from_system to the ChatMessage does also not work.

Why is it checking for text in the system message? Shouldnt it be checking for .content?

generator = GoogleAIGeminiChatGenerator(
    # model="gemini-exp-1206",
    # model="gemini-2.0-flash",
    model="gemini-2.0-flash-thinking-exp-1219",
    generation_config={
        "temperature": 0.5,
        "max_output_tokens": 5000
    }
)

from haystack.dataclasses import ChatMessage

system_message = ChatMessage.from_system(content="answer in german")
user_message = ChatMessage.from_user(content="How many r are in strawberry?")


messages = [system_message, user_message]
res = generator.run(messages=messages)

print(res)

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[11], [line 42](vscode-notebook-cell:?execution_count=11&line=42)
     [38](vscode-notebook-cell:?execution_count=11&line=38) user_message = ChatMessage.from_user(content="How many r are in strawberry?")
     [41](vscode-notebook-cell:?execution_count=11&line=41) messages = [system_message, user_message]
---> [42](vscode-notebook-cell:?execution_count=11&line=42) res = generator.run(messages=messages)
     [44](vscode-notebook-cell:?execution_count=11&line=44) print(res)

File d:\Development\kashrok\.venv\lib\site-packages\haystack_integrations\components\generators\google_ai\chat\gemini.py:293, in GoogleAIGeminiChatGenerator.run(self, messages, streaming_callback)
    [281](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:281) """
    [282](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:282) Generates text based on the provided messages.
    [283](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:283) 
   (...)
    [290](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:290)     - `replies`:  A list containing the generated responses as `ChatMessage` instances.
    [291](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:291) """
    [292](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:292) streaming_callback = streaming_callback or self._streaming_callback
--> [293](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:293) history = [self._message_to_content(m) for m in messages[:-1]]
    [294](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:294) session = self._model.start_chat(history=history)
    [296](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:296) new_message = self._message_to_part(messages[-1])

File d:\Development\kashrok\.venv\lib\site-packages\haystack_integrations\components\generators\google_ai\chat\gemini.py:293, in <listcomp>(.0)
    [281](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:281) """
    [282](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:282) Generates text based on the provided messages.
    [283](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:283) 
   (...)
    [290](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:290)     - `replies`:  A list containing the generated responses as `ChatMessage` instances.
    [291](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:291) """
    [292](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:292) streaming_callback = streaming_callback or self._streaming_callback
--> [293](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:293) history = [self._message_to_content(m) for m in messages[:-1]]
    [294](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:294) session = self._model.start_chat(history=history)
    [296](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:296) new_message = self._message_to_part(messages[-1])

File d:\Development\kashrok\.venv\lib\site-packages\haystack_integrations\components\generators\google_ai\chat\gemini.py:262, in GoogleAIGeminiChatGenerator._message_to_content(self, message)
    [260](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:260) elif message.is_from(ChatRole.SYSTEM) or message.is_from(ChatRole.ASSISTANT):
    [261](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:261)     part = Part()
--> [262](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:262)     part.text = message.text
    [263](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:263) elif message.is_from(ChatRole.FUNCTION):
    [264](file:///D:/Development/kashrok/.venv/lib/site-packages/haystack_integrations/components/generators/google_ai/chat/gemini.py:264)     part = Part()

AttributeError: 'ChatMessage' object has no attribute 'text'

[Amnah199]

Hi @JanB1989, this error might be arising due to an older version of Haystack installed. We recently refactored ChatMessage data class to include text variable.
I would recommend installing latest version of haystack. You can also use simple commands like:

# check current haystack version 
pip list | grep haystack-ai

# install latest haystack version 
pip install -U haystack-ai

[JanB1989]

Just to clarify, this is the new synthax (text instead of content).

system_message = ChatMessage.from_system(text="answer in german")
user_message = ChatMessage.from_user(text="How many r are in strawberry?")