Google ADK Python v2.10.0
Google ADK Python v2.10.0 adds experimental skill lifecycle controls, a MongoDB vector/hybrid search toolset, evaluation metrics for cost/latency, and expanded OpenAI reasoning model support alongside several breaking behavioral changes.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
Google ADK Python release v2.10.0 introduces skill lifecycle management (including ephemeral lifecycles and active skill limits gated behind ADK ENABLE SKILL LIFECYCLE=1 ), MongoDB toolsets with vector and hybrid search capabilities, and evaluation metrics for tracking call duration, token usage, and model call counts. Several behavior changes and deprecations were introduced: - BigQuery protected write mode rejects non-SELECT statements without a clear dry-run session destination (such as multi-statement scripts, CALL , and EXPORT DATA ), and retains sessions in process memory instead of session state. - OpenAIResponsesLlm ignores thinking level / thinking budget in favor of OpenAIGenerateContentConfig.effort . - AgentEvaluator.evaluate and evaluate eval set now raise a ValueError if zero test cases are evaluated. - Instruction templating no longer resolves ${var} and \{var} from state, preserving them as literal strings. - Legacy live-audio modules in google.adk.flows.llm flows raise deprecation warnings on import and should be updated to google.adk.live .
可执行摘要
ADK Python v2.10.0 introduces skill lifecycle management (feature-flagged), MongoDB toolsets, and duration/token evaluation metrics. Breaking behavioral changes include restricted BigQuery protected write execution, in-memory session scoping for BigQuery temporary tables, template syntax preservation for ${var} , deprecation warnings for live-audio imports, and ValueError exceptions when evaluations contain zero…
- Agent 实用度
- 82/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Set the environment variable ADK ENABLE SKILL LIFECYCLE=1 if utilizing experimental skill lifecycles or unload tools.
- Update live-audio imports from google.adk.flows.llm flows to google.adk.live to prevent DeprecationWarning failures in test suites running with -W error.
- Replace thinking config (thinking level, thinking budget) with OpenAIGenerateContentConfig.effort when using OpenAIResponsesLlm.
- Ensure test suites using AgentEvaluator.evaluate or evaluate eval set have non-empty evaluation case sets to avoid new ValueError exceptions.