Skip to main content

Glossary - ADK Terms and Concepts

🎯 Purpose: Comprehensive reference for Google Agent Development Kit (ADK) terminology and concepts used throughout the tutorials.

📚 Source of Truth: google/adk-python (ADK 1.15) + Official Google Documentation

A​

Agent​

A complete AI system powered by a Large Language Model (LLM) that can perform tasks through tools, maintain state, and interact with users. Agents are more than just LLMs - they include reasoning, tools, memory, and instructions.

See Also: Tutorial 01: Hello World Agent

Agent-to-Agent (A2A) Communication​

Protocol for agents to communicate and collaborate with each other, enabling distributed multi-agent systems.

See Also: Tutorial 17: Agent-to-Agent Communication

Agent Engine​

Google Cloud's managed service for deploying and scaling agents on Vertex AI, providing built-in scaling, monitoring, and version management.

See Also: Tutorial 23: Production Deployment

B​

Built-in Tools​

Pre-built tools provided by Google ADK for common operations like web search, location services, and code execution.

See Also: Tutorial 11: Built-in Tools & Grounding

C​

Callbacks​

Functions that execute at specific points in an agent's lifecycle (before/after agent runs, tool calls, etc.) for monitoring, guardrails, and control flow.

See Also: Tutorial 09: Callbacks & Guardrails

Context Window​

The maximum amount of text (measured in tokens) that an LLM can process at once. Exceeding this limit causes errors.

CopilotKit​

React component library for building AI chat interfaces that integrate with ADK agents.

E​

Evaluation​

Systematic testing and quality assessment of agent behavior using automated metrics and human review.

See Also: Tutorial 10: Evaluation & Testing

Events​

Structured logging system that tracks agent execution, state changes, tool calls, and errors for debugging and monitoring.

See Also: Tutorial 18: Events & Observability

F​

Function Tools​

Regular Python functions that agents can call to perform specific tasks. ADK automatically generates schemas from function signatures and docstrings.

See Also: Tutorial 02: Function Tools

G​

Gemini​

Google's family of multimodal large language models, including Gemini 1.5, Gemini 2.0, etc.

See Also: Tutorial 22: Model Selection

Grounding​

Connecting LLM responses to real-world data and facts through tools like web search, databases, and APIs to ensure accuracy.

See Also: Tutorial 11: Built-in Tools & Grounding

Guardrails​

Safety mechanisms and validation rules that prevent agents from performing harmful actions or generating inappropriate content.

See Also: Tutorial 09: Callbacks & Guardrails

L​

Large Language Model (LLM)​

AI models trained on vast amounts of text data that can understand and generate human-like text. Examples: Gemini, GPT-4, Claude.

Loop Agent​

Workflow agent that iteratively refines output through critic/refiner patterns until quality criteria are met.

See Also: Tutorial 07: Loop Agents

M​

Memory Service​

Persistent storage system for long-term agent memory, enabling agents to recall information across sessions.

See Also: Tutorial 08: State & Memory

Model Context Protocol (MCP)​

Standardized protocol for tool communication between agents and external services, enabling interoperability.

See Also: Tutorial 16: MCP Integration

Multi-Agent Systems​

Architectures where multiple specialized agents work together to accomplish complex tasks.

See Also: Tutorial 06: Multi-Agent Systems

O​

Observability​

The ability to monitor, debug, and understand agent behavior through logging, metrics, and tracing.

See Also: Tutorial 18: Events & Observability, Tutorial 24: Advanced Observability

OpenAPI Tools​

Tools automatically generated from OpenAPI/Swagger specifications, allowing agents to call REST APIs without manual coding.

See Also: Tutorial 03: OpenAPI Tools

Output Key​

Configuration that automatically saves an agent's response to session state for later retrieval.

See Also: Tutorial 08: State & Memory

P​

Parallel Agent​

Workflow agent that executes multiple sub-agents simultaneously for improved performance on independent tasks.

See Also: Tutorial 05: Parallel Processing

Planners​

Advanced reasoning components that help agents break down complex tasks and create execution plans.

See Also: Tutorial 12: Planners & Thinking

Production Deployment​

Strategies for deploying agents to production environments with scalability, reliability, and monitoring.

See Also: Tutorial 23: Production Deployment

R​

Runner​

ADK component that executes agents, manages state, and coordinates tool calls.

S​

Sequential Agent​

Workflow agent that executes sub-agents in order, where each step depends on the previous step's output.

See Also: Tutorial 04: Sequential Workflows

Session State​

Key-value storage that persists data within a conversation session but is discarded when the session ends.

See Also: Tutorial 08: State & Memory

State Management​

System for storing and retrieving data across agent interactions, with different scopes (session, user, app, temp).

See Also: Tutorial 08: State & Memory

Streaming​

Real-time response generation where the agent sends partial responses as they are generated, rather than waiting for completion.

See Also: Tutorial 14: Streaming & SSE

Server-Sent Events (SSE)​

HTTP standard for real-time communication from server to client, used for streaming agent responses.

See Also: Tutorial 14: Streaming & SSE

T​

Tool Context​

Object passed to tool functions containing state, session information, and execution context.

Tools​

Capabilities that extend agent functionality beyond LLM reasoning. Types include function tools, OpenAPI tools, MCP tools, and built-in tools.

See Also: Tools & Capabilities

V​

Vertex AI​

Google Cloud's machine learning platform that provides managed AI services including Gemini models and Agent Engine.

W​

Workflow Agents​

Agents that orchestrate other agents in structured patterns: sequential, parallel, and loop workflows.

See Also: Workflows & Orchestration

Quick Reference Tables​

Agent Types​

TypePurposeExample Use Case
LLM AgentFlexible reasoning and conversationCustomer support, content creation
Sequential AgentOrdered, dependent stepsBlog writing pipeline
Parallel AgentIndependent concurrent tasksResearch gathering
Loop AgentIterative refinementCode review and improvement

State Scopes​

PrefixScopePersistenceExample
(none)Current sessionSessionService dependentstate['topic']
user:All user sessionsPersistentstate['user:language']
app:All users/sessionsPersistentstate['app:settings']
temp:Current invocationNever persistedstate['temp:calc']

Tool Types​

TypeSourceExample
Function ToolsPython functionsCustom business logic
OpenAPI ToolsREST API specsWeather, news APIs
MCP ToolsMCP serversFilesystem, databases
Built-in ToolsGoogle ADKSearch, maps, code execution

Workflow Patterns​

PatternExecutionUse Case
SequentialOne after anotherAssembly line processes
ParallelAll at onceIndependent research tasks
LoopRepeat until criteria metQuality improvement cycles

Contributing to the Glossary​

This glossary is maintained alongside the ADK tutorials. When new concepts are introduced:

  1. Add the term with a clear definition
  2. Include "See Also" links to relevant tutorials
  3. Update related terms if needed
  4. Keep definitions concise but comprehensive

Last Updated: October 2025 ADK Version: 1.15