Agent Engine Live API based voice agent

[A-b-h-a-y-0-2]

Hello,
We are trying to deploy the voice agents on Agent Engine as it could help with concurrency. Will it be able to accommodate the live api based voice functionality?

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: Yes, you can deploy voice agents with live, streaming capabilities to Agent Engine. This functionality relies on the experimental bidi-streaming (live) feature in the ADK, which uses the Gemini Live API. You'll need to use a Gemini model that supports this API.

---

Hello! Thanks for your question.

The functionality you're referring to as "live api based voice functionality" is what the ADK refers to as "bidi-streaming (live)". It is an experimental feature that enables real-time, bidirectional streaming of audio, allowing for the creation of sophisticated voice agents. [1]

To implement this, you will need to use a Gemini model that supports the Gemini Live API. You can find more details about the models that support this in the documentation. [1]

Agent Engine is indeed a suitable choice for deploying and managing your voice agent, especially when you need to handle concurrency and scale. It's a fully managed service designed for this purpose and is available for Python agents. When you deploy your agent to Agent Engine, it will automatically use the VertexAiSessionService for managing conversational memory. [2]

For a comprehensive guide on building and deploying your first agent, you can refer to the Quickstart guide. [3]

Citations:
[1] Streaming - ADK: https://google.github.io/adk-docs/streaming/
[2] Deploy to Agent Engine: https://google.github.io/adk-docs/deploy/agent-engine/
[3] Quickstart: https://google.github.io/adk-docs/get-started/quickstart/