How do I install the Adk Streaming skill?

Run npx skillmds@latest add folken2/adk-streaming in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Adk Streaming safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Adk Streaming?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Adk Streaming free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Adk Streaming?

Folken2 (@folken2) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Adk Streaming by folken2 — Agent Skill

Name: Adk Streaming
Author: Folken2

Adk Streaming

Building voice and video agents with Gemini Live API — bidirectional streaming via WebSocket, LiveRequestQueue, run_live(), RunConfig, and audio/video patterns. Load when the user wants a voice agent, live agent, or real-time audio/video capabilities.

Folken2 6e7345a 3 files · 8.6 KB Updated Sep 17, 2026

File contents

ADK Streaming — Voice & Video Agents

Build real-time voice and video agents using ADK's bidirectional streaming with the Gemini Live API.

When to Use

The user explicitly asks for:

Voice agent / voice assistant
Live agent / real-time agent
Audio/video capabilities
Conversational agent with speech

Architecture

Browser (mic/cam) → WebSocket /ws/{user_id}/{session_id}
    ↓ upstream_task
LiveRequestQueue (text: send_content, audio: send_realtime)
    ↓
Runner.run_live() ←→ Gemini Live API (BIDI mode)
    ↓ downstream_task
WebSocket → Browser (audio playback + transcriptions)

Key Components

Component	Purpose
`LiveRequestQueue`	Async FIFO buffer between WebSocket and agent. One per session.
`Runner.run_live()`	Async generator consuming from queue, yielding events
`RunConfig(streaming_mode=StreamingMode.BIDI)`	Configures BIDI mode for Live API
`StreamingMode.BIDI`	WebSocket to Gemini Live API (audio/video)
`StreamingMode.SSE`	HTTP streaming to standard Gemini API (text only)

Template Integration

The generated template includes streaming support gated behind STREAMING_ENABLED=true:

streaming.py — WebSocket handler with upstream/downstream tasks
run_adk.py — conditional branch that builds Runner manually with LIVE_MODEL
static/test_client.html — minimal browser test client
config/llm.py — LIVE_MODEL config (Gemini model string, not LiteLLM)

Design Decisions for Streaming Agents

When designing a streaming agent:

Model: Must use Gemini directly (LIVE_MODEL), not OpenRouter/LiteLLM
API Key: Requires GOOGLE_API_KEY (not OPENROUTER_API_KEY)
Tools work during streams: Agent can call FunctionTools mid-conversation
VAD: Automatic by default — no manual activity signals needed
Transcription: Enable both input and output transcription for logging
Session resumption: Enable for connection recovery

Customization Points

When generating a streaming agent, customize these in streaming.py:

Voice: Add speech_config to RunConfig for voice selection
Language: Set language_code in speech config
VAD mode: Disable auto VAD for push-to-talk UIs
Video: Add video modality to response_modalities
Audio format: Default is PCM 16kHz input, 24kHz output

Load references for details on each customization.

References

Load streaming-patterns for RunConfig options, VAD modes, voice selection, and audio format details.
Load live-api-reference for LiveRequestQueue API, event types, and error handling.

---name: adk-streamingdescription: Building voice and video agents with Gemini Live API — bidirectional streaming via WebSocket, LiveRequestQueue, run_live(), RunConfig, and audio/video patterns. Load when the user wants a voice agent, live agent, or real-time audio/video capabilities.---# ADK Streaming — Voice & Video AgentsBuild real-time voice and video agents using ADK's bidirectional streamingwith the Gemini Live API.## When to UseThe user explicitly asks for:- Voice agent / voice assistant- Live agent / real-time agent- Audio/video capabilities- Conversational agent with speech## Architecture```Browser (mic/cam) → WebSocket /ws/{user_id}/{session_id} ↓ upstream_taskLiveRequestQueue (text: send_content, audio: send_realtime) ↓Runner.run_live() ←→ Gemini Live API (BIDI mode) ↓ downstream_taskWebSocket → Browser (audio playback + transcriptions)```## Key Components| Component | Purpose ||-----------|---------|| `LiveRequestQueue` | Async FIFO buffer between WebSocket and agent. One per session. || `Runner.run_live()` | Async generator consuming from queue, yielding events || `RunConfig(streaming_mode=StreamingMode.BIDI)` | Configures BIDI mode for Live API || `StreamingMode.BIDI` | WebSocket to Gemini Live API (audio/video) || `StreamingMode.SSE` | HTTP streaming to standard Gemini API (text only) |## Template IntegrationThe generated template includes streaming support gated behind `STREAMING_ENABLED=true`:1. `streaming.py` — WebSocket handler with upstream/downstream tasks2. `run_adk.py` — conditional branch that builds Runner manually with `LIVE_MODEL`3. `static/test_client.html` — minimal browser test client4. `config/llm.py` — `LIVE_MODEL` config (Gemini model string, not LiteLLM)## Design Decisions for Streaming AgentsWhen designing a streaming agent:1. **Model**: Must use Gemini directly (`LIVE_MODEL`), not OpenRouter/LiteLLM2. **API Key**: Requires `GOOGLE_API_KEY` (not `OPENROUTER_API_KEY`)3. **Tools work during streams**: Agent can call FunctionTools mid-conversation4. **VAD**: Automatic by default — no manual activity signals needed5. **Transcription**: Enable both input and output transcription for logging6. **Session resumption**: Enable for connection recovery## Customization PointsWhen generating a streaming agent, customize these in `streaming.py`:- **Voice**: Add `speech_config` to `RunConfig` for voice selection- **Language**: Set `language_code` in speech config- **VAD mode**: Disable auto VAD for push-to-talk UIs- **Video**: Add video modality to `response_modalities`- **Audio format**: Default is PCM 16kHz input, 24kHz outputLoad references for details on each customization.## References- Load `streaming-patterns` for RunConfig options, VAD modes, voice selection, and audio format details.- Load `live-api-reference` for LiveRequestQueue API, event types, and error handling.

Frequently asked questions