Google Agent Development Kit
AIOpen SourceOpen-source, code-first toolkit for building, evaluating, and deploying modular AI agents across models and environments.
-(0 Reviews)
Free Available
Starting from Free
Google Agent Development Kit
AIOpen Source
Open-source, code-first toolkit for building, evaluating, and deploying modular AI agents across models and environments.
-(0 Reviews)
Free Available
Starting from Free
About Google Agent Development Kit
Agent Development Kit (ADK) is an open-source, code-first toolkit from Google for building, testing, orchestrating, and deploying agentic applications. ADK provides Python (and Java) libraries, a developer Web UI, and runtime primitives that let developers define agent logic, tools, sessions, and evaluations as code. It is model-agnostic and deployment-agnostic — optimized for Google Gemini but compatible with other LLMs via adapters — and includes features for tool integration, context and session memory, multi-turn orchestration, artifact management, and evaluation pipelines. ADK is distributed under the Apache-2.0 license and is designed to make agent development feel like standard software engineering with installation via pip/maven and an emphasis on modularity, observability, and extensibility.
Key Features
Code-First Development: Define agents, tools, invocation logic, and test suites entirely in Python (with Java support) so agent behavior is authored and versioned like application code.
Model-Agnostic Execution: Run agents with Google Gemini models or any compatible LLM via adapters (e.g., LiteLLM), enabling developers to switch backends without rewriting agent logic.
Tool Integration and Extensibility: Expose arbitrary Python functions or external APIs as agent capabilities (tools) for data fetching, computation, or side-effectful actions integrated into agent workflows.
Orchestration & Session Management: Built-in runner and session primitives for managing multi-turn dialogues, invocation contexts, event streams, and long-running workflows across agents.
Contextual Memory & State: Native support for session memory and context tracking so agents maintain state across interactions and can reference prior events, artifacts, and conversation history.
Developer Web UI (ADK Web): A built-in web-based developer interface for building, debugging, inspecting traces, and iterating on agents with visualized runs and event logs.
Artifact & Evaluation Tools: First-class handling of file/artifact storage, code execution environments, and evaluation tooling for running test cases, measuring agent performance, and automating QA.
Deployment Agnostic & Observability: Designed to deploy across environments with integrations for tracing/tracing tools and support for instrumenting runs for monitoring and debugging.
Code-first Python SDK for defining agents, tools, runs, and sessions
Java SDK with Maven/Gradle artifacts for JVM integration
Model-agnostic design with optimization for Google Gemini and support for other LLMs (e.g., via LiteLLM)
Deployment-agnostic orchestration enabling simple to complex multi-agent workflows
Built-in developer web UI (ADK Web) for development and debugging
Session, memory, and context management for multi-turn interactions
Tool integration: plug Python functions or external APIs as agent capabilities
Artifact service for file read/write and managing runtime artifacts
Authentication and credential management helpers for secure external service access
Evaluation framework for testing agent behavior and measuring performance
CLI and FastAPI integration for running agents as services and streaming responses
Open-source license (Apache-2.0) with public docs, samples, and walkthroughs
Use Cases
Custom Assistants: Build a domain-specific content-assistant (e.g., support, documentation, or content generation) by wiring LLM prompts, tools for data retrieval, and session memory to maintain context across interactions.
Multi-Agent Workflows: Orchestrate coordinated multi-agent systems where agents have distinct roles (researcher, analyst, executor) and pass results through invocation contexts and sessions.
API & Tool Orchestration: Wrap external APIs or internal services as ADK tools to enable agents to fetch live data, run computations, or perform side effects (e.g., database updates, ticket creation).
Developer Debugging & Iteration: Use ADK Web to step through agent runs, inspect invocation traces, replay sessions, and iterate on logic and tool behavior during development.
Evaluation and Regression Testing: Create automated test suites and evaluation pipelines to validate agent output quality, compare LLM backends, and catch regressions before deployment.
Production Deployment: Package and deploy agents across environments (local, cloud, containerized) while leveraging ADK's session tracing, artifact management, and model-agnostic adapters for operational flexibility.
Build conversational assistants and chatbots with multi-turn memory and tool access
Compose multi-agent workflows to automate complex, multi-step tasks
Integrate external APIs or custom code as agent capabilities (data fetch, compute, orchestration)
Develop, debug, and iterate agents using the ADK Web developer UI
Run and serve agents as web services via FastAPI integration
Evaluate and test agent performance with built-in evaluation tooling
Deploy agents on Vertex AI or other deployment targets while retaining model-agnostic architecture
Screenshots
+
+
