Practical overview
When it fits
Use ADK when you want a Python-based, code-first toolkit spanning agent implementation, evaluation, and deployment.
Who it is for
- Python developers building AI agents
- Teams creating agent evaluation workflows
- Developers preparing deployable agent services
- Prototypers exploring agentic applications
Documented uses
- Build Python-based AI agents
- Create evaluation workflows for AI agents
- Develop deployable agent services
- Prototype agentic applications
Requirements & limits
The project is an accessible public GitHub repository, described as open source and licensed under Apache-2.0. Its platform context includes CLI, cloud, self-hosted, and server environments.
- Python is the stated primary language.
- The toolkit covers building, evaluation, and deployment rather than only agent prototyping.
- Its code-first design is intended to provide control over agent implementation and lifecycle.
- The repository is public, active, and licensed under Apache-2.0.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install ADK and create a basic Python agent
Install the stable ADK package and define a basic greeting agent in Python.
Before you start
- Python 3.10 or newer.
- A Python environment with pip available.
Steps
Install the stable package with the command below.
pip install google-adkCreate a Python module that imports
Agentfromgoogle.adkand defines an agent with a name, model, and instruction.For dependency constraints, download the constraints file matching the Python version, install with the command below using the matching filename, and remove the file afterward.
pip install google-adk -c constraints-3.10.txt
Expected outcome
ADK is installed and a basic `Agent` definition is available for further development.
Guide limitations
- The documented constraints example is for Python 3.10; use the corresponding documented constraints file for other supported Python versions.
- The source documents agent definition but does not document model credentials or configuration.
Run an ADK agent locally and evaluate the bundled sample
Use the documented local CLI or web UI entry points and run the repository’s bundled evaluation sample.
Before you start
- ADK must be installed.
- A local agent directory is required for the local run commands.
- The evaluation sample paths must be available from the repository root.
Steps
Run an interactive local agent with
adk run path/to/my_agent, or useadk web path/to/agents_dirfor the web UI.From the repository root, run the documented evaluation command targeting
contributing/samples/evaluation/home_automation_agentand its evaluation set.Pass the documented evaluation configuration file with
--config_file_path contributing/samples/evaluation/basic_criteria/eval_config.json.
Expected outcome
The documented local interface is started for an agent, and the bundled evaluation sample is invoked with its evaluation set and configuration.
Guide limitations
- The source does not document expected evaluation results or successful execution.
- The source does not provide the contents or required structure of a custom agent directory.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 30 September 2026
- Original website title
- google/adk-python · Official homepage title · retained in its source language