Real-time Multimodal Latency: Enhanced support for bidirectional voice and video via the Live API. This enables native audio reasoning for robotic and mobile agents. 4. Ecosystem Integration

Agent Identity: Unique digital IDs for agents to enforce the principle of least privilege in enterprise environments.

Next steps: Clarification may be needed to determine if "2.8.7" refers to a specific GitHub repository version, a CLI tool, or perhaps a typo for a different version like 2.5 or 3.x. Welcome to Google Cloud Next26

Chapter-Based Narrative Flow: This groups agent interactions into "Chapters" based on user intent and tool usage. This improves the coherence of long-running tasks.

Multimodal Embedding 2: A unified embedding space that supports text, image, video, audio, and PDF inputs simultaneously for more complex cross-modal retrieval. 3. Key Feature Benchmarks