DEV.co
AI Frameworks · google

adk-python

ADK (Agent Development Kit) is Google's open-source Python framework for building, evaluating, and deploying AI agents using a code-first approach. It provides a workflow runtime with graph-based execution, task delegation between agents, and local/web-based tooling for development and testing.

Source: GitHub — github.com/google/adk-python
20.5k
GitHub stars
3.7k
Forks
Python
Primary language
Apache-2.0
License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

FieldValue
Repositorygoogle/adk-python
Ownergoogle
Primary languagePython
LicenseApache-2.0 — OSI-approved
Stars20.5k
Forks3.7k
Open issues703
Latest releasev2.4.0 (2026-07-07)
Last updated2026-07-08
Sourcehttps://github.com/google/adk-python

What adk-python is

ADK 2.0 introduces a graph-based workflow execution engine supporting routing, fan-out/fan-in, loops, retry logic, and nested workflows alongside a Task API for structured agent-to-agent delegation. Built in Python 3.10+, it integrates with LLMs (e.g., Gemini) and provides CLI and web UI for local development and multi-agent orchestration.

Quickstart

Get the adk-python source

Clone the repository and explore it locally.

terminalbash
git clone https://github.com/google/adk-python.gitcd adk-python# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Multi-Agent Orchestration & Workflows

Building deterministic, graph-based workflows where multiple agents collaborate, delegate tasks, and handle branching logic with built-in retry and state management.

Agent Development & Local Testing

Iterative development of AI agents with interactive CLI and web UI, enabling rapid prototyping, debugging, and evaluation before deployment.

Task Delegation Systems

Implementing structured agent-to-agent task handoff with multi-turn conversation modes, human-in-the-loop approval, and mixed delegation patterns.

Implementation considerations

  • Plan migration from ADK 1.x if running production agents; sessions are forward-compatible to 1.28+ but breaking API changes require code refactoring.
  • Verify Python 3.10+ availability in target deployment environment; no older version support.
  • Configure LLM credentials and model selection upfront (e.g., Gemini-2.5-flash); model availability and billing model impact agent behavior and cost.
  • Design workflow graph structure early; complex nested workflows and large fan-out patterns may require optimization for latency and token usage.
  • Evaluate optional extensions (pip install google-adk[extensions]) to determine which integrations are needed for your agent tasks and tools.

When to avoid it — and what to weigh

  • Python < 3.10 Required — If your infrastructure is locked to Python 3.9 or earlier, this toolkit cannot be used without upgrading your runtime environment.
  • Simple Chatbot Only — For single-agent, stateless chatbot use cases, ADK's workflow and task infrastructure introduces unnecessary complexity; a lighter LLM library may suffice.
  • Breaking Change Constraints — If you are running ADK 1.x agents in production with strict backward compatibility requirements, ADK 2.0's API and schema changes require migration planning.
  • Offline-Only or Closed-Network Deployments — ADK's typical usage integrates with Google LLMs (e.g., Gemini) and cloud services; air-gapped or entirely local inference requires custom integration effort.

License & commercial use

Apache License 2.0 (SPDX: Apache-2.0). Permissive OSI-approved license allowing commercial use, modification, and redistribution with attribution and disclaimer of liability.

Apache 2.0 permits commercial use, but review your own legal terms regarding use of Google-maintained code, any bundled dependencies, and LLM API cost structures. Consult your counsel for production deployment in regulated industries.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

SignalAssessment
MaintenanceActive
DocumentationAdequate
License clarityClear
Deployment complexityModerate
DEV.co fitStrong
Assessment confidenceHigh
Security considerations

LLM API credentials (keys, model selection) must be secured via environment variables or secrets management. Workflow execution with external tools (agent tools defined by users) should validate inputs to prevent prompt injection. State and session data handling should follow data residency and privacy policies (Google servers). No explicit security audit or threat model documented.

Alternatives to consider

LangChain / LangGraph

Broader ecosystem, multi-LLM support (OpenAI, Anthropic, etc.), larger community. More flexible but also steeper learning curve; not Google-specific.

Semantic Kernel (Microsoft)

Azure/Copilot integration, plugins, enterprise support. Better for organizations already in Microsoft ecosystem; less workflow-native than ADK.

CrewAI

Role-based agent framework with simpler multi-agent setup. Lighter-weight alternative if you do not need complex workflow graphs or task APIs.

Software development agency

Build on adk-python with DEV.co software developers

Start with ADK's quick-start guide, explore sample workflows, and deploy locally. For production orchestration and integrations, contact our engineers.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

Related on DEV.co

Explore the category and the services that help you build with it.

adk-python FAQ

Can I use ADK with non-Google LLMs?
Not directly shown in provided documentation. Examples use Gemini; custom LLM integration requires extending ADK's model layer, which is not clearly documented.
What is the difference between Agent and Workflow?
Agent is a single AI entity with instructions, tools, and a model. Workflow is a graph-based orchestrator that composes multiple agents, handles routing, branching, and task delegation.
Is ADK 2.0 backward compatible with 1.x?
No. ADK 2.0 has breaking API and schema changes. Sessions from 2.0 are readable by ADK 1.28+ (backward-compatible), but upgrading from 1.x requires code migration.
How is state managed across workflow nodes?
Not fully detailed in provided excerpt. ADK mentions 'state management' in workflow runtime but explicit state passing, context, and persistence mechanisms require documentation review.

Software developers & web developers for hire

Need help beyond evaluating adk-python? DEV.co is a software development agency offering software development services and web development for teams of every size. Our software developers and web developers build custom software, web applications, APIs, and ai frameworks integrations — and maintain them long-term.

Ready to Build AI Agents?

Start with ADK's quick-start guide, explore sample workflows, and deploy locally. For production orchestration and integrations, contact our engineers.