adk-python v2.3.0: New features and bug fixes
sdkaipreviewdeprecationengineermediapython-agentic
feature patch
The google/adk-python library has been updated to version 2.3.0, introducing several new features such as enhanced caching, GCS integration, mTLS support, and expanded LLM capabilities with Gemma4 and Live API translation. This release also addresses numerous bugs across various modules including core functionality, evaluation, and agent interactions, improving stability and performance for developers using the ADK. The update aims to provide a more robust and feature-rich environment for building AI-powered applications with Python.
- →add create_http_options to ContextCacheConfig for cache creation timeout , closes #4703
- →add GCS first party toolset to ADK integrations
- →Add log_level option for adk run CLI
- →Add mTLS support in AgentRegistry client
- →add request timeout to load_web_page
Features (20) ›
- add create_http_options to ContextCacheConfig for cache creation timeout , closes #4703
- add GCS first party toolset to ADK integrations
- Add log_level option for adk run CLI
- Add mTLS support in AgentRegistry client
- add request timeout to load_web_page
- core migrate core and CLI to enterprise parameters
- Create GEPARootAgentOptimizer
- eval expose user_simulator_config in generate_responses
- gemma4 support Gemma4 in Gemini
- integrations Add E2BEnvironment for remote sandbox workspaces
- labs add experimental Antigravity SDK agent wrapper
- lazily initialize the Vertex AI client in GCPSkillRegistry, and fix the import path for vertexai
- live Handle input transcription differently for Gemini Live 3.1 models
- live support Live API translation config in RunConfig
- Migrate McpToolset to AsyncAuthorizedSession for mTLS support
- plugins ADK 2.0 minimum producer cut for the BigQuery Agent Analytics plugin
- report cached token counts for Anthropic and OpenAI models
- telemetry support per-request OpenTelemetry configuration
- Update gcp_auth client UI to support Remote Agents
- utils add GOOGLE_GENAI_USE_ENTERPRISE env var with deprecation fallback
Enhancements (2) ›
- flows skip async-rearrange when no function_responses
- test Speed up unit test suite via parallelism and dedup
Fixes (79) ›
- a2a Preserve execution metadata in final events
- a2a render HITL interrupt when prompt is in a data part
- a2a set final=True for error TaskStatusUpdateEvent in event_converter
- a2a suppress part_metadata in Vertex AI mode
- add a2a as a required dependency for agent_engine deployment
- adk propagate exceptions from run_node in standalone mode
- adk propagate isolation_scope to prevent history filtering loops
- agents await cancelled tasks in _merge_agent_run_pre_3_11 to prevent aclose() RuntimeError
- api-registry to fetch all services , closes #5478
- artifacts Support nested API names
- auth handle missing client-credentials scopes safely , closes #5345
- avoid UserWarning in _build_response_log when response has funct…
- call to sanitize schema for complex union types
- ci add repository check to prevent workflows from running on forks
- ci Resolve missing sqlalchemy error in adk_release_analyzer
- cli Serialize LiteLlm graph models safely
- conformance normalize tool declarations in replay verification
- Default subagents to chat mode in build_node
- deps Require otel google-genai instrumentor >=0.7b1 for genai 2.x
- eval handle failed inference results without invocations
- eval handle unevaluated final response v2 results
- eval include function-call events in invocation_events when skip_summarization is set
- eval preserve custom eval metadata
- Extract grounding_metadata from Live API server_content
- Fix silent dead end when conditional routes are unmatched
- Fix typing for create_client in mcp_session_manager
- fix vertex_ai_session_service crashing when Agent Engine passes full resource names instead of short session IDs
- flows terminate invocation at tool-level EUC
- Format files to fix pre-commit failures
- Format the files
- gate pr-triage secrets on same-repository pull_request_target
- handle missing agent name in readonly context
- improve error message when beautifulsoup4/lxml not installed for load_web_page , closes #4852
- live history_config rejection on Vertex/Enterprise Live sessions
- live propagate output token count in live API usage metadata
- log diagnostics for empty or unparseable rubric auto-rater output , closes #5732
- make DatabaseSessionService visible in API docs , closes #4331
- Mock google.auth.default in test_fast_api.py
- models pass NOT_GIVEN to Anthropic when no system_instruction , closes #5318
- models surface error when model returns STOP with empty content
- models surface error when model returns STOP with empty content
- Move google-cloud-parametermanager to optional dependencies
- Only send grounding_metadata for 3.1 live at the end of each turn
- otel Handle empty contents in experimental semconv
- planners allow BuiltInPlanner subclasses to override process_planning_response
- planners keep all leading parallel function calls
- plugins write BigQuery analytics rows when invocation agent is None
- preserve empty GCS text artifacts
- Preserve event details when output is delegated
- preserve function call ids for litellm models , closes #2621
- prevent compaction from orphaning function responses
- prevent ReDoS in code block extraction
- propagate model_version and other metadata in streaming responses
- remove developer notes from transfer_to_agent docstring
- remove live event buffering in runner
- remove the issue/PR analyze and fix agent workflows
- remove the issue/PR triage and fix agent workflows
- Reset retry attempt counter on successful connection
- restore GitHub-only changes dropped during v2 bring-over
- sessions Further fixes for DatabaseSessionService
- sessions honor zero recent events in database service
- sessions Prevent MissingGreenlet after append_event with asyncpg
- Set role='model' for request_input event Content
- skills enforce utf-8 encoding when materializing skill files on Windows
- skip crewai test on ImportError for pytest 9.1 compatibility
- Stop interpolating release analyzer workflow inputs into shell commands
- support non-Latin text in InMemoryMemoryService search
- surface MALFORMED_FUNCTION_CALL so on_model_error can recover
- tests skip bash tool tests on Windows
- tools dereference draft-07 definitions in MCP tool schemas
- tools handle missing 'request' key in AgentTool.run_async fallb…
- update model to gemini-3.5-flash in session_state_agent sample
- use correct 'content' key in sandbox code executor input files
- utils Preserve decorated type for @experimental and @working_in_progress
- utils Preserve decorated type for @experimental and @working_in_progress
- workflow Preserve explicit single-turn contents
- workflow Prevent replay divergence hang in sequence barrier
- workflow Prevent Shared InvocationContext branch mutation
- workflow Prevent silent drain of routed nodes in wait_for_output
Maintenance (22) ›
- Add diversion logic based on the auth provider resource name
- Implement the auth provider using Agent Identity Credentials service
- Move the IamConnectorCredential service depedency to a seperate file
- otel Add pure functions for constructing stable and experimental semconv logs
- Remove unused imports across src
- Separate PR analysis from triage for automation
- telemetry change agent and tool execution duration metrics from milliseconds to seconds
- add beginner explanation for single agent example
- add PyPI, Python version, downloads, and docs badges to README
- add unit guides for event.py, request_input.py and update adk-unit-guide skill
- add unit guides for task mode
- Align Python version to 3.10+ and update README badges
- clarify context cache min_tokens gating and 4096-token minimum
- Fix ADK release analyzer session db saving error
- fix formatting in multi-agent sample READMEs
- fix triaging agent sample typo
- openapi improve docs for session model
- remove stale -b v2 flag from clone command in CONTRIBUTING.md
- skills fix broken refs in adk-workflow skill
- update llms.txt and remove build script
- Upgrade ADK release analyzer agent to use gemini-3.1-pro-preview model
- Add async and concurrency guidelines to ADK Style Guide
Read the original announcement →
https://github.com/google/adk-python/releases/tag/v2.3.0
Related releases
- Pydantic-AI v2.9.0: Security fix, GPT-5.6 models, usage command Pydantic AI Releases ·
- LangGraph 1.2.7: Fixes for delta overwrites and UUID generation LangGraph Releases ·
- pydantic-ai v2.1.0: Anthropic web tools, serialization, and bug fixes Pydantic AI Releases ·
- pydantic-ai v2.5.1: Bug Fixes and Model Compatibility Pydantic AI Releases ·
- crewAI 1.15.0: Conversational Flow Tracking and Declarative Flow Loading CrewAI Releases ·
- OpenAI Agents Python Library v0.17.8: Enhancements and Fixes OpenAI Agents SDK (Python) Releases ·