Skip to content
Agent Guides

Google ADK (Python)

Google’s Agent Development Kit. Build agents on Gemini (and any LLM via google-genai / LiteLlm) with graph-based Workflow orchestration, MCP client + server, A2A protocol, and native GCP services (Vertex AI, Cloud Run, Agent Engine).

Install

Terminal window
pip install google-adk

Version

v2.9.0 · September 2026 · Python 3.10+

Best for

GCP-native workloads, Gemini-first agents, Vertex AI Agent Engine.

Zero → Hero

Section titled “Zero → Hero”
1 · Introduction What ADK is, why Google built it · 10 min
2 · Core fundamentals Install, auth, first agent · 25 min
3 · Agents LlmAgent, Parallel, Loop, Sequential · 25 min
4 · Workflows Graph orchestration with @node · 30 min
5 · Tools FunctionTool, AgentTool, built-ins · 30 min
6 · Callbacks & plugins Interception points · 20 min
7 · Runner & sessions App, Runner, session services · 25 min
8 · Memory & artifacts Long-term memory + file storage · 20 min
9 · MCP & A2A Tool servers + agent handoff · 20 min
10 · Structured output Pydantic validation · 15 min
11 · Advanced Python Planners, custom tools, auth, deployment · 30 min
12 · Evaluation AgentEvaluator, EvalCase, EvalSet, metrics · 25 min
13 · IAM & security IAM roles, workload identity · 20 min
14 · Production Deployment, scaling, monitoring · 30 min
15 · Class & API reference Consolidated, source-verified reference for every ADK class across 12 subject areas — replaces the old 48-volume deep-dive series · source-verified against google-adk 2.7.1 (2.8.0 additions in steps 16 and 17)
16 · Class deep dives — v2.8.0 10 source-verified deep dives for 2.8.0: LiveRequestQueue, ExecuteBashTool, ToolConfirmation, App, EventsCompactionConfig, LlmEventSummarizer, PreloadMemoryTool, ToolboxToolset, GcsArtifactService, exit_loop + get_user_choice_tool
17 · Class deep dives Vol. 3 — v2.8.0 10 more source-verified deep dives: RunConfig, BuiltInPlanner, PlanReActPlanner, BasePlugin, ReflectAndRetryToolPlugin, ContextFilterPlugin, VertexAiMemoryBankService, VertexAiCodeExecutor, ConversationScenarios + LlmBackedUserSimulatorConfig, SaveFilesAsArtifactsPlugin

Jump to a topic

Section titled “Jump to a topic”
LlmAgent
Workflow graphs
Tools & built-ins
Callbacks & plugins
Runner & sessions
Memory & artifacts
MCP & A2A
IAM examples
Advanced Python
Evaluation

Reference

Section titled “Reference”
Agents LlmAgent, Parallel, Loop, Sequential (deprecated).
Workflows Workflow, @node, Edge, JoinNode.
Tools FunctionTool, AgentTool, BaseTool, BaseToolset, MCP.
Callbacks & plugins All 13 plugin hooks + agent callbacks.
Runner & sessions Runner, App, session services, RunConfig.
Memory & artifacts MemoryService + ArtifactService implementations.
MCP & A2A McpToolset, RemoteA2aAgent, to_a2a().
Evaluation AgentEvaluator, EvalCase, EvalSet, PrebuiltMetrics.
Comprehensive guide Single-page technical reference.
Advanced Python Planners, custom tools, auth, deployment.
IAM examples
Recipes
Diagrams
Production guide
Class deep dives v2.8.0 LiveRequestQueue, ExecuteBashTool, ToolConfirmation, App, EventsCompactionConfig, LlmEventSummarizer, PreloadMemoryTool, ToolboxToolset, GcsArtifactService, exit_loop + get_user_choice_tool
Class deep dives Vol. 3 v2.8.0 RunConfig, BuiltInPlanner, PlanReActPlanner, BasePlugin, ReflectAndRetryToolPlugin, ContextFilterPlugin, VertexAiMemoryBankService, VertexAiCodeExecutor, ConversationScenarios + LlmBackedUserSimulatorConfig, SaveFilesAsArtifactsPlugin

Install, minimal example

Section titled “Install, minimal example”
import asyncio
from google.adk.agents import LlmAgent
from google.adk.runners import InMemoryRunner


agent = LlmAgent(
    name="assistant",
    model="gemini-2.5-flash",
    instruction="You are a helpful assistant.",
)


async def main():
    runner = InMemoryRunner(agent=agent, app_name="demo")
    await runner.session_service.create_session(app_name="demo", user_id="u1", session_id="s1")
    events = await runner.run_debug("What is 15 + 27?", user_id="u1", session_id="s1")
    print(events[-1].content.parts[0].text)


asyncio.run(main())

Ready? Start Zero → Hero Step 1 →

Changelog

Section titled “Changelog”

Compact, version-level history. Per-symbol details for the current release live in the Class & API Reference above; older per-volume changelog detail has been superseded and lives in git history.

VersionDateSummary
1.18.0Nov 2025Initial documented version.
1.31.0Apr 2026A2A protocol, MCP client + server, Web UI overhaul (Graph View, execution traces), Vertex AI Agent Engine Sandbox, Firestore session support.
2.0.0 (beta → GA)Apr – May 2026Workflow graph orchestration ships; SequentialAgent/ParallelAgent/LoopAgent deprecated in its favor. Plugin system (BasePlugin, 11+ hooks). MCP sampling/resources/auth schemes. to_a2a() / RemoteA2aAgent. Vertex AI Memory Bank + RAG. Built-in code executors (local, Vertex AI, container, GKE, Agent Engine sandbox). GA stable release confirmed May 20, 2026 — --pre no longer required.
2.1.0May 2026RunConfig gains tool_thread_pool_config, context_window_compression, get_session_config; BaseNode schema fields (state_schema/input_schema/output_schema) formalized; Context.add_memory().
2.2.0Jun 2026AgentEngineSandboxComputer, AgentRegistry, GcpAuthProvider, ApigeeLlm; live-audio session classes stabilized (AudioCacheManager, AudioTranscriber, TranscriptionManager); ParameterManagerClient/SecretManagerClient.
2.3.0Jun 2026Credential-exchange pipeline (BaseCredentialExchanger, OAuth2CredentialExchanger), SessionStateCredentialService, workflow error taxonomy (NodeInterruptedError/NodeTimeoutError/DynamicNodeFailError), conformance-testing infrastructure, ToolConnectionAnalyzer.
2.4.0Jul 2026ManagedAgent (Managed Agents API), DaytonaEnvironment, ReplayManager, injectable platform.time/uuid/thread for deterministic tests, PubSubCredentialsConfig/PubSubToolConfig.
2.5.0Jul 2026SkillRegistry/SkillToolset, AgentOptimizer/Sampler ABCs, GEPARootAgentOptimizer, OpenAILlm (labs), LangGraphAgent, AutoFlow vs SingleFlow pipeline split.
2.6.1 – 2.6.3Aug 2026App/EventsCompactionConfig/ResumabilityConfig refinements, ToolConfirmation, McpInstructionProvider, comprehensive RunConfig/StreamingMode/ToolThreadPoolConfig, SpannerToolset, CachePerformanceAnalyzer.
2.7.1Aug 21, 2026Documentation consolidation. The 48 sequential “class deep dive” volumes (covering versions 2.1.0 → 2.6.3) have been merged, deduplicated, and re-verified as one Class & API Reference section on the comprehensive guide, organized into 12 subject areas. Notable import-path corrections confirmed against the installed 2.7.1 package: EventsCompactionConfig moved to the private google.adk.apps._configs (no longer re-exported from google.adk.apps); every google.adk.a2a class requires a private submodule import; GCS/Firestore session, memory, and artifact integrations relocated to google.adk.integrations.*; LoopAgent/ParallelAgent/SequentialAgent remain functional but are confirmed @deprecated in favor of Workflow.
2.8.0Aug 31, 2026New deep-dive page covering 10 source-verified classes: LiveRequestQueue (bidirectional streaming), ExecuteBashTool/BashToolPolicy (secure bash with HITL gate), ToolConfirmation (human-in-the-loop protocol), App (top-level container), EventsCompactionConfig (token-threshold + sliding-window), LlmEventSummarizer (LLM-driven session summarizer), PreloadMemoryTool (proactive memory injection), ToolboxToolset (Cloud Toolbox / MCP Toolbox), GcsArtifactService (GCS artifact storage), and exit_loop/get_user_choice_tool (control-flow built-ins).

Note: Other languages: Go · TypeScript.