What is bidi-streaming in Google ADK?

Bidirectional (bidi) streaming allows for simultaneous, low-latency communication where the agent can process and respond while the user is still providing input, supporting natural interruptions.

Can I switch response modalities mid-session?

No, Google ADK requires you to select one response modality (either TEXT or AUDIO) per session; switching requires initializing a new session.

How does session resumption work in this skill?

It provides configurations for SessionResumptionConfig, which allows the agent to automatically reconnect and maintain conversation context across connection timeouts.

Does this skill support both Google AI Studio and Vertex AI?

Yes, it includes patterns for toggling between Google AI Studio and Vertex AI using environment variables without requiring code changes.

What are streaming tools in the context of ADK?

Streaming tools are defined as async generators that yield intermediate results over time, enabling the agent to provide continuous updates for long-running tasks.

ADK Real-Time Streaming Patterns

Name: ADK Real-Time Streaming Patterns
Author: vanman2024

byvanman2024

0•

Data Science & ML

Configures Google ADK bidirectional streaming to build low-latency, multimodal AI agents with real-time voice and video capabilities.

This skill provides comprehensive implementation patterns for Google's Agent Development Kit (ADK) to enable bidirectional (bidi) streaming between users and AI agents. It covers critical configurations like LiveRequestQueue management, multimodal response handling, and session resumption for connection stability. It is essential for developers building Gemini-powered voice assistants, real-time video analysis tools, and sophisticated multi-agent systems that require seamless context preservation and asynchronous tool execution.

Key Features

01Multimodal support for simultaneous text, audio, and video processing

02Bidirectional streaming configuration for real-time AI interactions

03Streaming tool patterns using async generators for continuous data output

04Multi-agent session transfer for seamless stateful handoffs

050 GitHub stars

06Advanced session management with automatic resumption and context compression

Use Cases

01Implementing low-latency monitoring tools that stream live data updates

02Developing real-time voice and video AI agents using Gemini Live API

03Building complex multi-agent workflows that maintain conversation context across sessions

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add vanman2024/ai-dev-marketplace streaming-patterns

For use in Claude.ai and ChatGPT

Download Skill