[Bug] Gemini agent with memory errors with empty formatted_messages

[jeffbryner]

Description

Was trying out the fixes for gemini in v1.1.4 and got a strange error once hooking up memory with sqlite to an agent.

Steps to Reproduce

Add tools, memory and storage to an agent and set everything to True:
create_user_memories=True,
create_session_summary=True, # seems especially troublesome
update_user_memories_after_run=True,
update_session_summary_after_run=True,

agent = Agent(
    model=model_flash,
    tools=[DuckDuckGoTools(), CsvTools(csvs=[imdb_csv])],
    markdown=True,
    show_tool_calls=True,
    telemetry=False,
    monitoring=False,
    instructions=[
        "You have a local CSV file named imdb.csv with IMDB movie data",
        "You can run a query with this csv file to answer questions.",
        "You can also search the internet with DuckDuckGo.",
    ],
    storage=SqliteAgentStorage(table_name="chat_agent", db_file=agent_storage),
    # Adds the current date and time to the instructions
    add_datetime_to_instructions=True,
    # Adds the history of the conversation to the messages
    add_history_to_messages=True,
    # Number of history responses to add to the messages
    num_history_responses=15,
    memory=AgentMemory(
        db=SqliteMemoryDb(db_file="tmp/agent_memory.db"),
        create_user_memories=True,
        create_session_summary=True,  # troublesome
        update_user_memories_after_run=True,
        update_session_summary_after_run=True,
        classifier=MemoryClassifier(model=model_flash),
        summarizer=MemorySummarizer(model=model_pro),
        manager=MemoryManager(model=model_flash),
    ),
)

Agent Configuration (if applicable)

Agent using the most recent gemini fixes.

Expected Behavior

Expected memory/session summaries to work

Actual Behavior

Odd error kept occurring:

    return self.get_client().models.generate_content(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/jeffbryner/development/agno-playground/.venv/lib/python3.12/site-packages/google/genai/models.py", line 4672, in generate_content
    response = self._generate_content(
               ^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/jeffbryner/development/agno-playground/.venv/lib/python3.12/site-packages/google/genai/models.py", line 3807, in _generate_content
    request_dict = _GenerateContentParameters_to_vertex(
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/jeffbryner/development/agno-playground/.venv/lib/python3.12/site-packages/google/genai/models.py", line 1180, in _GenerateContentParameters_to_vertex
    for item in t.t_contents(
                ^^^^^^^^^^^^^
  File "/Users/jeffbryner/development/agno-playground/.venv/lib/python3.12/site-packages/google/genai/_transformers.py", line 321, in t_contents
    raise ValueError('contents are required.')
ValueError: contents are required.

Environment

• Agno Version: v1.1.4
• External Dependency Versions: duckduck go tools, csv tools

Possible Solutions (optional)

A simple exit if empty will fix the issue for me locally:

        if not formatted_messages:
            return GenerateContentResponse()
        try:
            return self.get_client().models.generate_content(
                model=self.id,
                contents=formatted_messages,
                **request_kwargs,
            )
...

in each of the functions that gather formatted messages before calling the model

• invoke
• invoke_stream
• ainvoke
• ainvoke_stream

Additional Context

It appears that the

formatted_messages, system_message = self._format_messages(messages)

which is made before calling the gemini model

        try:
            return self.get_client().models.generate_content(
                model=self.id,
                contents=formatted_messages,
                **request_kwargs,
            )

can result in empty 'formatted_messages' .

When I added some logging I got logs showing calls with no messages:

INFO     Sending messages to Gemini model: []

[pritipsingh]

Hey @jeffbryner thanks for raising this. @dirkbrnd has a pr out to fix this. will be fixed in the next release!

[dirkbrnd]

Hi!
This was fixed in release 1.1.5