Google released ADK Python v1.30.0 on April 13 at 23:16 UTC, adding native Gemma 4 model support, Auth Provider integration for agent registries, and artifact streaming in A2A (agent-to-agent) protocol events. The release also patches a credential leakage vulnerability in the Agent Registry.

What Shipped

The v1.30.0 changelog lists nine features and eight bug fixes. The headline additions:

Gemma 4 support. Google’s open multimodal model family can now run natively inside ADK multi-agent pipelines. The commit closes issue #5156, a community request that had been open for weeks.

Auth Provider for Agent Registry. A new integration (commit f2c68eb) adds structured credential management when agents register with and discover other agents in a shared registry. This matters for teams running multi-agent systems where agents need authenticated access to each other’s capabilities.

A2A artifact streaming. Agents communicating through Google’s A2A protocol can now include artifacts from the artifact service in A2A events using a provided interceptor (commit e63d991). Previously, agents could only exchange text and status updates in agent-to-agent calls. Now they can pass live artifacts: files, data objects, or other structured outputs.

Live avatar support. ADK agents can now render live avatars (commit a64a8e4), a capability aimed at customer-facing agent deployments where visual presence matters.

BigQuery tools promoted to stable. The BigQuery toolset, previously in preview, is now a first-class stable component (commit abcf14c).

Parameter Manager integration. A new integration (commit b0715d7) centralizes configuration management for ADK agents.

Security Fixes

The release patches a credential leakage vulnerability in the Agent Registry (commit e3567a6) and adds path traversal validation for user_id and session_id parameters (commit cbcb5e6), closing issue #5110.

Framework Context

ADK Python has shipped 30 point releases in 2026 alone. InfoWorld’s hands-on review, published the same day as the release, confirmed the framework now supports custom service registration, session rewind capabilities, and workflow graph visualization in the web UI. The review positions ADK against Amazon Bedrock AgentCore, Azure AI Foundry Agents, Databricks Agent Bricks, LangChain/LangGraph, and CrewAI.

The ADK is available in four languages (Python, Go, Java, TypeScript) with a shared web development UI. It supports both LLM-based and deterministic workflow agents, multi-agent orchestration, and deployment to Vertex AI Agent Engine or any containerized environment.

The Auth Problem for Multi-Agent Systems

The Auth Provider addition addresses a real gap. As agent registries grow, the question of how Agent A authenticates with Agent B becomes a production blocker. v1.29.0, released on April 9, had already added auth scheme support for MCP toolsets in the Agent Registry. v1.30.0 extends this to a general Auth Provider pattern, suggesting Google sees credential management as foundational infrastructure for agent ecosystems, not an afterthought.

For builders running multi-agent systems on ADK, the upgrade path is straightforward: pip install --upgrade google-adk pulls v1.30.0 from PyPI.