Google releases ADK 2.14.0 as agent-to-agent protocol A2A goes stable, adds Teams support

Google has released ADK Python 2.14.0, the latest version of its open-source agent development framework (dated 4 November, published on GitHub on 7 November). Execution environments, the A2A agent-to-agent protocol, evaluation and core runtime modules graduate from experimental to stable, and a new TeamsRunner connects ADK agents to Microsoft Teams chats and channels.

A2A goes stable, letting agents built on different frameworks hand off tasks

The A2A (Agent-to-Agent) protocol, hosted by the Linux Foundation, is designed to let agents built on different frameworks discover each other and hand off tasks. With the graduation, developers can generate an A2A agent card from the command line with adk generate_agent_card. Sandbox environments including BaseEnvironment, LocalEnvironment, Daytona and E2B also become stable.

Multi-run evaluation and faster long workflows

The release adds MultiRunConfig, which runs each evaluation case multiple times in parallel and combines the scores, and ToolArgsMatcher, which compares tool-call arguments semantically. It also adds an OpenAI Decisions integration and vector retrieval plus TTL expiry for MongoDbMemoryService. Google says a 30-node workflow running 200 turns is 23% faster with 9% lower peak memory, but it has not published the benchmark hardware or baseline version, so the figure is Google's own claim.

In total, 26 modules and 16 toolset and integration features graduate to stable, so developers can build on them in production without experimental warnings.

(Image above: official Google ADK artwork; image source: Google ADK.)