Agent Development Kit (ADK) 2.14.0 released

A major release graduating experimental features to stable, adding multi-run evaluation and semantic argument matching, and introducing Microsoft Teams and OpenAI Decisions integrations.

What changed in 2.14.0

• Execution environments, A2A, evaluation, and core runtime modules are now stable; large file reads and writes stream in chunks to avoid memory issues.
• Multi-run evaluation runs each case multiple times concurrently with MultiRunConfig and --parallelism, and ToolArgsMatcher compares tool arguments using exact, case-insensitive, numeric, or semantic checks.
• New TeamsRunner integrates agents with Microsoft Teams, OpenAIDecisionsClassifier gates workflows with the OpenAI Decisions API, and MongoDbMemoryService adds vector search.
• Long multi-turn workflows are 23% faster with 9% lower peak memory; 26 auth, environment, and app modules plus 16 toolset features graduate from experimental.
• AgentEvaluator accepts agents built at runtime, and --eval_storage_uri stores evaluation files outside read-only agent directories.

Check before you update

• Enum tool parameters now receive enum members instead of strings; code comparing against strings or using string methods must use .value or switch to StrEnum.
• VertexAiSessionService.create_session accepts only display_name, ttl, expire_time, and labels; other config keys now raise ValueError.
• RunSkillScriptTool with SkillToolset environment no longer accepts command; scripts receive args, short_options, and positional_args instead, and only .py, .sh, and .bash scripts are supported.

Summarised by AgentGid from Google's official release notes on GitHub, as reported there; read the original before relying on it. Price, status and release dates come from our daily data.