NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI, #1243 most downloaded on PyPI
Agent Development Kit
Latest stable
2.14.1
Last release 9 days ago
12 Nov 2026
Ships on a steady schedule
a new release about every 2 weeks
Nearly every release is documented
notes for 59 of the last 60 stable releases
3 versions withdrawn
withdrawn after publishing
2 years old
100 releases · first in 2025
Drag across the ruler to compare, or tap a release to open it.
Add chart generation and artifact loading to data agent
Your coding agent can read these notes before it upgrades. Set up the MCP server →
This release introduces v2.0.0 General Availability (GA) of the Google Agent Development Kit (ADK), establishing production-grade foundations for mult
This release introduces v2.0.0 General Availability (GA) of the Google Agent Development Kit (ADK), establishing production-grade foundations for multi-agent workflows and advanced dynamic agent collaboration.
Security Fix: Resolved a potential RCE vulnerability related to nested YAML configurations (2b54c4ac).
LlmAgent instances (d864917f).Workflow Orchestration: Added Workflow(BaseNode) graph orchestration implementation, support for lazy scan deduplication and resume for dynamic nodes,
add agent name validation to prevent arbitrary module imports
Introduces two major capabilities:
Introduces two major capabilities:
a2a: Port event action allowlist, agent card description source and resume flattening to v1
live: forward safety_settings from generate_content_config to the Live API
live: Support asynchronous non-blocking tools
Update custom gemini llm connection logic to be used for all 3_x models, not just 3.1
exit connection cleanly on expected GoAway signal in bidi streaming (v1)
interactions: support for GenAI SDK upgraded to 2.9+
remove live event buffering in runner
live: Handle input transcription differently for Gemini Live 3.1 models
Nothing published for this version
Fix bug where grounding metadata in Gemini 3.1 live was being silently discarded
Fix bug where grounding metadata in Gemini 3.1 live was being silently discarded
a2a: add support for persistent task stores
apigee: allow injecting credentials into ApigeeLlm
block RCE vulnerability via nested YAML configurations in ADK
gen_ai.tool_definitions to gen_ai.tool.definitions (029b87d)output_schema docstring to reflect support for tools and output_schema together (e1e652d)block RCE vulnerability via nested YAML configurations in ADK
New Live UI: Redesigned the real-time chat and testing interface for a cleaner, more responsive user experience.
_EvalMetricResultWithInvocation.expected_invocation Optional for conversation_scenario support (#5215) (a4c9387)send_client_content to send conversation history (67dc2eb)send_tool_response for function responses (70c5fc8)Fix credential leakage vulnerability in Agent Registry
TaskStatusUpdateEvent for ADK events with no output parts but with event.actions (dcc485b)Add auth scheme/credential support to MCP toolsets in Agent Registry
SecretManagerClient (19ac679)gemini-3.1-flash-live-preview model (8082893)live: support live for gemini-3.1-flash-live-preview model
a2a: add lifespan parameter to to_a2a() (0f4c807), closes #4701
required for Pydantic BaseModel parameters in FunctionTool (c5d809e), closes #4777find_context_parameter (22fc332)Update eval extras to Vertex SDK package version with constrained LiteLLM upperbound
Exclude compromised LiteLLM versions from dependencies pin to 1.82.6
add protection for arbitrary module imports
Use valid dataplex OAuth scope for BigQueryToolset
Rolling back change to fix issue affecting LlmAgent creation due to missing version field
Introduce A2A request interceptors in RemoteA2aAgent
[Core]
types.SchemaUnion as output_schema in LLM Agent (63f450e)[Models]
[Web]
[Observability]
gen_ai.tool.definitions to experimental semconv (4dd4d5e)gen_ai.client.inference.operation.details event (19718e9)[Tools]
[Optimization]
[Integrations]
[A2A]
Add intra-invocation compaction and token compaction pre-request
[Core]
[Models]
[Tools]
load_skill_from_dir() method (9f7d5b3)_meta field with OpenTelemetry propagator (bcbfeba)[Evals]
[Services]
[A2A]
A2aAgentExecutor (87fcd77)[Auth]
[Integrations]
[Web]
input_stream parameter name and its annotation to decide whether it's a streaming tool that accept input stream (d56cb41)DatabaseSessionService (fbe9ecc)thinking_config in generate_content_config in example agent (c6b1c74)Fix pickling lock errors in McpSessionManager
Add a demo for the simple prompt optimizer for the optimization interface
[Core]
--auto_create_session flag to adk api_server CLI (40c15d0)add_events_to_memory facade for event-delta (59e8897)adk conformance test command (43c437e)[Models]
[Tools]
[Web]
/health and /version endpoints to ADK web server (25ec2c6)/run endpoint to Runner.run_async (d2dba27)Add back deprecated eval endpoint for web until we migrate
Disable save_input_blobs_as_artifacts deprecation warning message for users not setting it
tool_context instead of callback_context (fe82f3c)[Core]
[Models]
[Evals]
[Tools]
McpToolset, OpenAPIToolset, and other toolsets (798f65d)get_tools calls (ee873ca)VertexAiSearchTool (585ebfd)get_auth_config method to toolset to expose authentication requirements (381d44c)McpToolset for users to access MCP resources (8f7d965)[Services]
asyncio loop handling and test cleanup (00aba2d)[Live]
[Observability]
opentelemetry-instrumentation-google-genai (e87a843)session_db_kwargs for SQLite session services (ce07cd8)MutualTLSChannelError by adding pyopenssl dependency (125bc85)update_timestamp_tz property to StorageSession (666cebe)execution_result_delimiters have None type element (a16e3cc)save_input_blobs_as_artifacts deprecation warning message for users not setting it (c34615e)pypika>=0.50.0 to project.toml to support crewai on Python 3.12+ (e8f7aa3)opentelemetry-api and opentelemetry-sdk (706a6dd)pool_pre_ping by default for non-SQLite database engines (da73e71)DatabaseSessionService when storage update time is later than the in-memory session object (1063fa5)read_resource to return original contents (ecce7e5)ClientSession (131fbd3)adk eval help message (3bcd8f7)will_continue for streaming function calls (2220d88)gemma_llm.py (574ec43)MCPToolset to McpToolset in unit tests (2770012, 4341839)BigQueryAgentAnalyticsPlugin is shut down after each test (c0c98d9)RestApiTool (288c2c4)mypy (32f9f92)unittests.sh script and update CONTRIBUTING.md (025b42c)Breaking: Use OpenTelemetry for BigQuery plugin tracing, replacing custom ContextVar implementation
ContextVar implementation (ab89d12)[Core]
@experimental decorator from AgentEngineSandboxCodeExecutor (135f763)--disable_features CLI option to override default feature enable state (53b67ce)otel_to_cloud flag to adk deploy agent_engine command (21f63f6)is_computer_use field to agent information in adk-web server (5923da7)thinking_config in generate_content_config (e162bb8)DataPart metadata and ADK events (1133ce2)--enable_features CLI option to override default feature enable state (79fcddb)[Tools]
BigQueryAgentAnalyticsPlugin to ensure pending log events are written to BigQuery (9579bea)base_retrieval_tool, load_artifacts_tool, and load_memory_tool declarations when the feature is enabled (69ad605)IntegrationConnectorTool declaration when the feature is enabled (2ed6865)ClientSession in a single task in McpSessionManager (cce430d)RestApiTool declaration when the feature is enabled (a5f0d33)[Evals]
[Observability]
generate_content {model.name} spans and logs for non-Gemini inference and when opentelemetry-inference-google-genai dependency is missing (935c279)[Integrations]
TraceManager asynchronous safety, enrich BigQuery plugin logging, and fix serialization (a4116a6)[Live]
function_call and function_response events during live streaming (7b25b8f)authlib and mcp dependency versions (7955177)LITELLM_MODE to PRODUCTION before importing LiteLLM to prevent implicit .env file loading (215c2f5)None inferences in eval results (7d4326c)json.dumps for error messages in SSE events (6ad18cc)None content (ed2c3eb)'NoneType' object is not iterable error (7db3ce9)tool.name (ec6abf4)self._auth_config inside BaseAuthenticatedTool to access authentication headers in McpTool (d4da1bb)LoadArtifactsTool (fdc98d5)log_level to uvicorn in web and api_server commands (38d52b2)NOT_FOUND error when fetching Vertex AI sessions (75231a3)httpx client closure during event pagination (b725045)adk_request_input event from content list (295b345)ToolboxToolset to use toolbox-adk and align validation (7dc6adf 277084e)DebugLoggingPlugin to record human-readable debugging logs (8973618)gemini-2.5-flash (fd2c0f5)migration_runner lazily within the migrate command (905604f)Add back adk migrate session CLI (8fb2be2).
Move and enhance the deprecation warning for the plugins argument in "_validate_runner_params" to the beginning of the function (43270bc).
[Core]
[Evals]
[Tools]
[Services]
[A2A]
adk create command (742c926).override_feature_enabled to override the default feature enable states (a088506).Update FastAPI and Starlette to fix CVE-2025-62727 (ReDoS vulnerability)
[Interactions API Support]
Agent(
model=Gemini(
model="gemini-3-pro-preview",
use_interactions_api=True,
),
name="...",
description="...",
instruction="...",
)
see samples for details
[Services]
add_session_to_memory to CallbackContext and ToolContext to explicitly save the current session to memory (7b356dd)[Plugins]
[A2A]
[Tools]
[Models]
build_function_declaration_with_json_schema to use pydantic to generate json schema for FunctionTool (51a638b)invocation_id and new_message in run_async (de841a4)Add enum constraint to agent_name for transfer_to_agent
[Core]
[Models]
[Tools]
[Web]
base_llm_flow.py where, in Bidi-streaming (live) mode, the multi-agent structure causes duplicated responses after tool calling. (cf21ca3)Fix: Fixes DeprecationWarning when using send method
[Core]
id and custom_metadata fields to MemoryEntry (4dd28a3)[Evals]
[Integrations]
[Services]
[Models]
[Tools]
[Observability]
jsonschema dependency for Agent Builder config validation (0fa7e46)event in remote_a2a_agent.py (744f94f)FunctionDeclaration without a required attribute (93aad61)async for to loop through event iterator to get all events in vertex_ai_session_service (9211f4c)google-cloud-aiplatform version in agent engine requirements (e15e19d)load_artifacts tool (c485889)max_query_result_rows in BigQuery tool config (fd33610)cleanup_unused_files.py (2dea573)google.cloud.storage in GCSArtifactService (999af55)live, Client and _transformers in google.genai (22c6dbe)plugin_reflect_tool_retry sample (01bac62)Visual workflow designer for agent creation
[ADK Visual Agent Builder]
[Core]
agent_to_a2a (156d235)[Models]
[Integrations]
[Services]
[Tools]
[Evals]
[Observability]
[Live]
format field from LiteLLM content objects (489c39d)Add a service registry to provide a generic way to register custom service implementations to be used in FastAPI server. See short instruction here.
[Core]
[Evals]
[Integrations]
[Observability]
[Services]
[Tools]
bypass_multi_tools_limit option to GoogleSearchTool and VertexAiSearchTool (9a6b850, 6da7274)ReflectAndRetryToolPlugin to support hallucinating function calls (f51380f)[UI]
additionalProperties from Gemini schemas (307896a)App instances returned by agent_loader.load_agent (847df16)2 and above for gemini-builtin tools (0df6759)bypass_multi_tools_limit to False for GoogleSearchTool and VertexAiSearchTool (6da7274)ReflectAndRetryToolPlugin (9b8a4aa)ReflectAndRetryToolPlugin (9b8a4aa)Implementation of LLM context compaction
dry_run functionality to BigQuery execute_sql tool (960eda3)ReflectRetryToolPlugin to reflect from errors and retry with different arguments when tool errors (e55b894)VertexAiSearchTool built-in tool with other tools in the same agent (4485379)adk web options for custom logo (822efe0)App object when loader returns an App (d5c46e4)pydantic validation errors when uploading files (e680063)max_output_tokens for the agent builder (2e2d61b)App instances returned by agent_loader.load_agent (847df16)Fix the deployment failure for Agent Engine
Adds plugin to save artifacts for issue #2176
Support explicit context caching auto creation and lifecycle management.
Usage: App(root_agent=..., plugins=..., context_cache_config=...)
Support static instruction that won't change, put at the beginning of the instruction. Static instruction support inline_data and file_data as contents. Dynamic instruction moved to the end of LlmRequest, increasing prefix caching matching size.
Usage:
LlmAgent(model=...,static_instruction =types.Content(parts=...), ... )
raw_mcp_tool method in McpTool to provide direct access to the underlying MCP tool (6158075)columns instead of modifying it in place (aef1ee9)after_agent_callback (b9735b2)Fix logging issues with RemoteA2aAgent 0c1f1fa
Allow users to pass their own agent card to to_a2a method a1679da
custom_metadata to DatabaseSessionService fb009d8
ALTER TABLE events ADD COLUMN custom_metadata JSON; to migrate existing database tables.thought handling in contents.py and refactors its unit tests a30851ethought field handling in _planning.py fe8b37blocal_eval_service and runners 64f11a6local_eval_service and runners 5b465fdpyproject.toml ca5f7f1after_run callbacks 7720616App class and improve Runner initialization tests fc90ce9[Tools] Add the ask_data_insights tool for natural language queries on BigQuery data 47b88d2
from_config class method in BaseToolset 2dd432cexecute_sql for various write modes self contained 167182bget_table_schema and fix the key usage info 81a53b5adk deploy cloud_run as Cloud Run args 6806dealitellm-test due to breaking change in dep library of extension extra
[Agent Config] 🌟 NEW FEATURE: Support using config file (YAML) to author agents in addition to python code. See the documentation for details.
build_image option for adk deploy cloud_run CLI (c843503)ADK_SUPPRESS_EXPERIMENTAL_FEATURE_WARNINGS as environment variable to suppress experimental warnings (4afc9b2)Field for Agent configs, so that the generated AgentConfig.json json schema can carry field description (5b999ed)openai dependency version, based on correct OPENAI release (bb8ebd1)[Tools] Add Spanner toolset (breaking change to BigQueryTool, consolidating into generic GoogleTool)
print_detailed_results param to AgentEvaluator.evaluate (7e08808)[Live] Implement Live Session Resumption
LlmRequest.LiveConnectConfig field default to a factory (74589a1)LlmRequest.config's default value to be types.GenerateContentConfig() (041f04e)One column per month.