google/adk-python v1.17.0

on GitHub

1.17.0 (2025-10-22)

Features

• [Core]

  • Add a service registry to provide a generic way to register custom service implementations to be used in FastAPI server. See short instruction here. (391628f)
  • Add the ability to rewind a session to before a previous invocation (9dce06f)
  • Support resuming a parallel agent with multiple branches paused on tool confirmation requests (9939e0b)
  • Support content union as static instruction (cc24d61)

• [Evals]

  • ADK cli allows developers to create an eval set and add an eval case (ae139bb)

• [Integrations]

  • Allow custom request and event converters in A2aAgentExecutor (a17f3b2)

• [Observability]

  • Env variable for disabling llm_request and llm_response in spans (e50f05a)

• [Services]

  • Allow passing extra kwargs to create_session of VertexAiSessionService (6a5eac0)
  • Implement new methods in in-memory artifact service to support custom metadata, artifact versions, etc. (5a543c0)
  • Add create_time and mime_type to ArtifactVersion (2c7a342)
  • Support returning all sessions when user id is none (141318f)

• [Tools]

  • Support additional headers for Google API toolset (ed37e34)
  • Introduces a new AgentEngineSandboxCodeExecutor class that supports executing agent-generated code using the Vertex AI Code Execution Sandbox API (ee39a89)
  • Support dynamic per-request headers in MCPToolset (6dcbb5a)
  • Add bypass_multi_tools_limit option to GoogleSearchTool and VertexAiSearchTool (9a6b850, 6da7274)
  • Extend ReflectAndRetryToolPlugin to support hallucinating function calls (f51380f)
  • Add require_confirmation param for MCP tool/toolset (78e74b5)

• [UI]

  • Granular per agent speech configuration (409df13)

Bug Fixes

• Returns dict as result from McpTool to comply with BaseTool expectations (4df9263)
• Fixes the identity prompt to be one line (7d5c6b9)
• Fix the broken langchain importing caused their 1.0.0 release (c850da3)
• Fix BuiltInCodeExecutor to support visualizations (ce3418a)
• Relax runner app-name enforcement and improve agent origin inference (dc4975d)
• Improve error message when adk web is run in wrong directory (4a842c5)
• Handle App objects in eval and graph endpoints (0b73a69)
• Exclude additionalProperties from Gemini schemas (307896a)
• Overall eval status should be NOT_EVALUATED if no invocations were evaluated (9fbed0b)
• Create context cache only when prefix matches with previous request (9e0b1fb)
• Handle App instances returned by agent_loader.load_agent (847df16)
• Add support for file URIs in LiteLLM content conversion (85ed500)
• Only exclude scores that are None (998264a)
• Better handling the A2A streaming tasks (bddc70b)
• Correctly populate context_id in remote_a2a_agent library (2158b3c)
• Remove unnecessary Aclosing (2f4f561)
• Fix pickle data was truncated error in database session using MySql (36c96ec)

Improvements

• Improve hint message in agent loader (fe1fc75)
• Fixes MCPToolset --> McpToolset in various places (d4dc645)
• Add span for context caching handling and new cache creation (a2d9f13)
• Checks gemini version for 2 and above for gemini-builtin tools (0df6759)
• Refactor and fix state management in the session service (8b3ed05)
• Update agent builder instructions and remove run command details (89344da)
• Clarify how to use adk built-in tool in instruction (d22b8bf)
• Delegate the agent state reset logic to LoopAgent (bb1ea74)
• Adjust the instruction about default model (214986e)
• Migrate invocation_context to callback_context (e2072af)
• Correct the callback signatures (fa84bcb)
• Set default for bypass_multi_tools_limit to False for GoogleSearchTool and VertexAiSearchTool (6da7274)
• Add more clear instruction to the doc updater agent about one PR for each recommended change (b21d0a5)
• Add a guideline to avoid content deletion (16b030b)
• Add an sample agent for the ReflectAndRetryToolPlugin (9b8a4aa)
• Improve error message when adk web is run in wrong directory (4a842c5)
• Add an sample agent for the ReflectAndRetryToolPlugin (9b8a4aa)
• Add span for context caching handling and new cache creation (a2d9f13)
• Disable the scheduled execution for issue triage workflow (bae2102)
• Correct the callback signatures (fa84bcb)

Documentation

• Format README.md for samples (0bdba30)
• Bump models in llms and llms-full to Gemini 2.5 (ce46386)
• Update gemini_llm_connection.py - typo spelling correction (e6e2767)
• Announce the first ADK Community Call in the README (731bb90)