added tool handling & triggered reconnection

[omChauhanDev]

Please describe the changes in your PR. If it is addressing an issue, please reference that as well.

Fixes #3549

When LLMSetToolsFrame was processed, the code called _update_settings() without the required argument, causing a TypeError. Since Gemini Live does not currently support mid-session tool updates, replaced the broken call with a warning log.

Before :

After :

[codecov]

Codecov Report

❌ Patch coverage is 0% with 1 line in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/pipecat/services/google/gemini_live/llm.py	0.00%	1 Missing ⚠️

Files with missing lines	Coverage Δ
src/pipecat/services/google/gemini_live/llm.py	19.72% <0.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[aconchillo]

We might not want to do this for now, reconnects are bad and slow. I believe we will wait until Gemini supports it. In the meantime we probably might want to remove the update_settings and probably add a warning instead saying that this is not supported in Gemini Live.

@kompfner has been looking into this and can comment.

[tropxy]

what I did as a workaround was to monkey patch the Gemini class locally

class GeminiLiveVertexLLMService(BaseGeminiLiveVertexLLMService):
    """Patched Gemini Live service that fixes bugs in the base implementation.

    Fixes:
    1. _update_settings bug: https://github.com/pipecat-ai/pipecat/blob/main/src/pipecat/services/google/gemini_live/llm.py#L953
       The base class calls _update_settings() without the required 'settings' argument.
    2. Transcription timeout task not being properly awaited when cancelled.
    """

    async def process_frame(self, frame: Frame, direction: FrameDirection):
        """Process frames with fix for LLMSetToolsFrame handling."""
        # Fix the bug: pass current settings when handling LLMSetToolsFrame
        if isinstance(frame, LLMSetToolsFrame):
            await self._update_settings(self._settings)
            # Continue processing through parent for other frame handling
            return

        # All other frames go through normal processing
        await super().process_frame(frame, direction)

    async def cancel_task(self, task, timeout: float = 1.0):
        """Override to properly await cancelled transcription timeout tasks."""
        if task and not task.done():
            task.cancel()
            try:
                # Properly await the cancelled task to avoid RuntimeWarning
                await asyncio.wait_for(task, timeout=timeout)
            except (asyncio.CancelledError, asyncio.TimeoutError):
                # Expected when task is cancelled
                pass
            except Exception as e:
                logger.debug(f"Error cancelling task: {e}")

[kompfner]

I have an open PR that adds the following (per the changelog entry):

- Added support to Gemini Live (`GeminiLiveLLMService`) for programmatically swapping tools or editing context at runtime; now you can use `LLMMessagesAppendFrame`, `LLMMessagesUpdateFrame`, `LLMMessagesTransformFrame`, and `LLMSetToolsFrame` with Gemini Live, like you would with text-to-text services. Note that this new functionality only works if you're using `LLMContext` and `LLMContextAggregatorPair` rather than the deprecated `OpenAILLMContext` and associated aggregators.

So we probably won't need this PR when that one lands.

Good catch, though, that the LLMSetToolsFrame handling in Gemini Live was never quite working as intended 🤦 .

Do note, however, that—as @aconchillo pointed out—a reconnect is necessary to make tool or conversation history changes at runtime. It's not too slow, but it might be noticeable (though not always).

[omChauhanDev]

make sense, i will close this one then