google__adk-docs/tools/feature-matrix/start.md

ADK Features

!!! note: This content is autogenerated by Gemini, based on our documentation. It is a best effort to generate a feature matrix, but may not be 100% accurate as features can outpace documentation, or Gemini might inaccurately generate content.

Feature Support Matrix

The following matrix shows which ADK features are currently supported in each language runtime. Support is determined by the availability of code samples in the official documentation.

Feature Category	Feature	Python (adk-python)	Java (adk-java)
Core Agent Architecture	LLM Agents	Supported	Supported
	Multi-Agent Systems	Supported	Supported
	Workflow: Sequential Agents	Supported	Supported
	Workflow: Loop Agents	Supported	Supported
	Workflow: Parallel Agents	Supported	Supported
	Custom Agents	Supported	Supported
	Agent Config (YAML)	Supported	No
Tooling & Integrations	Function Tools	Supported	Supported
	Long-Running Function Tools	Supported	Supported
	Agents-as-a-Tool	Supported	Supported
	Built-in: Google Search	Supported	Supported
	Built-in: Code Execution	Supported	Supported
	Built-in: GKE Code Executor	Supported	Not Supported
	Built-in: Knowledge Engine	Supported	Not Supported
	Built-in: Agent Search	Supported	Not Supported
	Built-in: BigQuery, Spanner, Bigtable	Supported	Not Supported
	Google Cloud Tools (API Hub, etc.)	Supported	Not Supported
	MCP Tools	Supported	Supported
	Tool Confirmation	Supported	Not Supported
	Tool Authentication	Supported	Not Supported
Runtime & Context	Session & State Management	Supported	Supported
	Memory	Supported	Not Supported
	Artifacts	Supported	Supported
	Callbacks	Supported	Supported
	Plugins	Supported	Not Supported
Advanced Capabilities	Streaming (Gemini Live API Toolkit)	Supported	Supported
	Streaming Tools	Supported	Not Supported
	Multi-Model: LiteLLM	Supported	Not Supported
	Multi-Model: Direct Anthropic	Not Supported	Supported
Development Lifecycle	Evaluation	Supported	Not Supported
	Observability (AgentOps, Arize, etc.)	Supported	Not Supported
	Deployment: Agent Runtime	Supported	No
	Deployment: Cloud Run	Supported	Supported
	Deployment: GKE	Supported	Not Supported
	Agent2Agent (A2A) Protocol	Supported	Not Supported

Details

Based on the documentation, ADK delivers a comprehensive set of features for building, evaluating, and deploying sophisticated AI agents. Here is a summary of the key capabilities:

Core Agent Architecture

• LLM Agents: The primary "thinking" unit of ADK, powered by a Large Language Model for reasoning, decision-making, and tool use.
• Multi-Agent Systems: Build complex applications by composing multiple, specialized agents in a hierarchy, enabling delegation and collaboration.
• Workflow Agents: Orchestrate the execution flow of other agents using deterministic patterns:
  • Sequential Agents: Run sub-agents one after another.
  • Parallel Agents: Run sub-agents concurrently.
  • Loop Agents: Run sub-agents in an iterative loop.
• Custom Agents: Implement unique, non-LLM based orchestration logic by extending the BaseAgent class for ultimate flexibility.
• Agent Config (YAML): Build agents declaratively using YAML configuration files, minimizing the need for boilerplate code.

Tooling and Integrations

• Function Tools: Extend agent capabilities by creating custom tools from standard functions.
• Long-Running Function Tools: Handle asynchronous tasks that take significant time to complete without blocking the agent.
• Agents-as-a-Tool: Use an entire agent as a callable tool within another agent.
• Built-in Tools: Leverage pre-built capabilities like Google Search, Agent Search, and secure code execution.
• Google Cloud Tools: Seamlessly connect to Google Cloud services like API Hub, Application Integration, and databases via MCP Toolbox.
• MCP (Model Context Protocol) Tools: Use and expose tools via the open MCP standard.
• Tool Confirmation: Implement human-in-the-loop workflows by requiring user confirmation before a tool executes.
• Tool Authentication: Securely manage credentials for tools that access protected APIs.

Runtime and Context Management

• Session & State: Manage conversational context with a robust session service that tracks history (Events) and maintains a working memory (State) for each interaction.
• Memory: Enable agents to recall information across multiple sessions using a long-term, searchable knowledge store.
• Artifacts: Allow agents to save, load, and manage versioned files and binary data (e.g., images, PDFs) associated with a session or user.
• Callbacks & Plugins: Hook into the agent's execution lifecycle to implement logging, guardrails, caching, and other cross-cutting concerns.

Advanced Capabilities

• Streaming (Gemini Live API Toolkit): Build real-time, interactive experiences with native support for bidirectional streaming of text, audio, and video.
• Streaming Tools: Create tools that can stream intermediate results back to the agent for real-time monitoring and reaction.
• Multi-Model Support: Flexibly use different LLMs (Gemini, GPT, Claude, etc.) within your agents, powered by integrations like LiteLLM.

Development Lifecycle

• Evaluation: Systematically assess agent performance by evaluating both the final response quality and the step-by-step execution trajectory.
• Observability & Logging: Gain deep insights into agent behavior with detailed logging and integrations with observability platforms like MLflow, AgentOps, Arize AX, Cloud Trace, and Weave.
• Deployment: Deploy your agents to a variety of environments, including serverless (Cloud Run), Kubernetes (GKE), and the fully managed Agent Runtime.
• Agent2Agent (A2A) Protocol: Build distributed, multi-agent systems where agents running as separate services can communicate and collaborate using the open A2A standard.