Add Gemini Model Provider

[mkmeral]

Add Gemini Live API Support for Bidirectional Streaming

Description

This PR adds support for Google's Gemini Live API as a bidirectional streaming model provider, enabling real-time audio conversations with native audio input/output, image/video input, and automatic transcription.

Key Features

Gemini Live Model Provider (gemini_live.py)

• Uses official google-genai SDK for robust WebSocket communication
• Native audio streaming with 16kHz input and 24kHz output
• Real-time audio transcription (both input and output)
• Image/video frame input support for multimodal conversations
• Automatic VAD-based interruption handling
• Tool calling integration
• Message history support

Enhanced Bidirectional Streaming

• Added ImageInputEvent type for sending images/video frames
• Added TranscriptEvent type for audio transcriptions (separate from text output)
• Extended BidirectionalAgent.send() to accept text, audio, and image inputs
• Updated abstract BidirectionalModelSession interface with send_image_content()

Test Suite Enhancements

• Updated test to support both Gemini Live and Nova Sonic
• Added camera capture for real-time video frame streaming (1 FPS)
• Demonstrates audio + video multimodal interaction
• Falls back to Nova Sonic if no Gemini API key provided

Implementation Details

The implementation follows the same architectural patterns as Nova Sonic:

• Provider-agnostic event conversion
• Clean separation between session management and model interface
• Simplified configuration - all Gemini Live API parameters pass through directly
• Proper async/await patterns with context manager for connection lifecycle

Configuration Example

from strands.experimental.bidirectional_streaming.models.gemini_live import GeminiLiveBidirectionalModel

model = GeminiLiveBidirectionalModel(
    model_id="gemini-2.5-flash-native-audio-preview-09-2025",
    api_key="your-api-key",
    params={
        "response_modalities": ["AUDIO"],
        "input_audio_transcription": {},   # Enable input transcription
        "output_audio_transcription": {},  # Enable output transcription
    }
)

Related Issues

Documentation PR

Type of Change

New feature

Testing

How have you tested the change?

• Tested real-time audio conversations with Gemini Live API
• Verified audio transcription (input and output) works correctly
• Tested image/video frame streaming from camera
• Verified tool calling integration
• Tested message history support
• Confirmed interruption handling via VAD
• Verified fallback to Nova Sonic when no API key provided
• Ran hatch fmt for code formatting

Test Environment

• Python 3.12+
• Dependencies: google-genai, pyaudio, opencv-python, pillow
• Tested with GOOGLE_AI_API_KEY environment variable

Files Changed

1. New: src/strands/experimental/bidirectional_streaming/models/gemini_live.py (501 lines)
2. Modified: src/strands/experimental/bidirectional_streaming/agent/agent.py - Added image input support
3. Modified: src/strands/experimental/bidirectional_streaming/models/bidirectional_model.py - Added abstract send_image_content() method
4. Modified: src/strands/experimental/bidirectional_streaming/models/novasonic.py - Added stub for image input (not supported)
5. Modified: src/strands/experimental/bidirectional_streaming/types/bidirectional_streaming.py - Added ImageInputEvent and TranscriptEvent types
6. Modified: src/strands/experimental/bidirectional_streaming/tests/test_bidirectional_streaming.py - Enhanced test with Gemini Live and camera support

Verify that the changes do not break functionality or introduce warnings in consuming repositories: agents-docs, agents-tools, agents-cli

• I ran hatch run prepare

Checklist

• I have read the CONTRIBUTING document
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[mehtarac]

suggestion: I think for testing purposes i would prefer a new test file dedicated for the gemini model provider.