Complete Google ADK CLI Command List

The current Google ADK Python CLI command tree includes api_server, conformance, create, deploy, eval, eval_set, migrate, optimize, run, test, and web. Your installed version may differ, so verify it locally with adk --help. Official ADK CLI reference

1. Display ADK help

adk --help

Shows every top-level command available in your installed ADK version.

For help with a specific command:

adk COMMAND --help

Example:

adk web --help

---

2. Display the installed ADK version

adk --version

Confirms which ADK CLI version is active.

---

3. Create a new agent project

adk create APP_NAME

Example:

adk create my_agent

Creates a new directory containing a prepopulated agent template.

Useful options:

adk create my_agent --model=MODEL_NAME

adk create my_agent --api_key=API_KEY

adk create my_agent \
  --project=PROJECT_ID \
  --region=REGION

Available options:

• --model: Model for the root agent
• --api_key: Google AI API key
• --project: Google Cloud project for Vertex AI
• --region: Google Cloud region for Vertex AI

Avoid placing API keys directly in shell history; environment variables or secret management are safer.

---

4. Run an agent in the terminal

Interactive mode:

adk run path/to/my_agent

Single-query mode:

adk run path/to/my_agent "Hello"

This runs the agent directly in the terminal without starting a browser.

Useful options:

adk run path/to/my_agent --save_session

adk run path/to/my_agent --resume=SESSION_FILE.json

adk run path/to/my_agent --replay=REPLAY_FILE.json

adk run path/to/my_agent --state='{"language":"English"}'

adk run path/to/my_agent --timeout=30s

adk run path/to/my_agent --in_memory

adk run path/to/my_agent --jsonl

Important options:

• --save_session: Save the session when exiting
• --session_id: Set the saved session ID
• --resume: Continue a saved session
• --replay: Replay queries from a JSON file
• --state: Provide initial state as JSON
• --timeout: Limit the duration of a turn
• --in_memory: Avoid persistent session storage
• --jsonl: Produce machine-readable JSONL output
• --default_llm_model: Set the default model
• --session_service_uri: Select session storage
• --artifact_service_uri: Select artifact storage
• --memory_service_uri: Select memory storage

---

5. Start the ADK development web interface

adk web

From the parent directory containing your agents:

adk web .

Specify the agents directory:

adk web path/to/agents

Specify a port:

adk web --port=8080 path/to/agents

Allow a frontend origin:

adk web \
  --allow_origins=https://example.com \
  path/to/agents

Enable debugging and automatic reload:

adk web \
  --log_level=DEBUG \
  --reload \
  --reload_agents \
  path/to/agents

Important options:

• --host: Binding address; default is 127.0.0.1
• --port: Server port
• --allow_origins: Allowed CORS origins
• -v or --verbose: Debug logging
• --log_level: Logging level
• --reload: Restart the server after code changes
• --reload_agents: Reload changed agents
• --url_prefix: Serve behind a path such as /adk
• --logo-text: Customize the UI logo text
• --logo-image-url: Customize the UI logo image
• --a2a: Enable the Agent-to-Agent endpoint
• --trace_to_cloud: Export Cloud Trace data
• --otel_to_cloud: Export OpenTelemetry data
• --default_llm_model: Supply a default model
• --extra_plugins: Enable additional plugins
• --trigger_sources: Enable sources such as Pub/Sub or Eventarc
• --session_service_uri: Configure session storage
• --artifact_service_uri: Configure artifact storage
• --memory_service_uri: Configure memory storage
• --use_local_storage: Store local ADK data under .adk

adk web is intended for development and testing—not as a production web interface.

---

6. Start the ADK API server

adk api_server

Specify the agents directory:

adk api_server path/to/agents

Specify the host and port:

adk api_server \
  --host=127.0.0.1 \
  --port=8000 \
  path/to/agents

Allow requests from a frontend:

adk api_server \
  --allow_origins=https://example.com \
  path/to/agents

Automatically create missing sessions:

adk api_server \
  --auto_create_session \
  path/to/agents

Include the web interface:

adk api_server \
  --with_ui \
  path/to/agents

This exposes agents through REST endpoints for connection to a website, mobile application, backend, or test client. By default, it runs at http://localhost:8000. ADK API Server documentation

It supports most of the server options available to adk web, including:

• --host
• --port
• --allow_origins
• --log_level
• --reload
• --reload_agents
• --a2a
• --url_prefix
• --trigger_sources
• --session_service_uri
• --artifact_service_uri
• --memory_service_uri
• --trace_to_cloud
• --otel_to_cloud

---

Deployment commands

7. Display deployment options

adk deploy --help

ADK currently provides deployment subcommands for:

• Cloud Run
• Agent Engine
• Google Kubernetes Engine

---

8. Deploy to Cloud Run

adk deploy cloud_run \
  --project=PROJECT_ID \
  --region=REGION \
  path/to/my_agent

With an explicit service name:

adk deploy cloud_run \
  --project=PROJECT_ID \
  --region=REGION \
  --service_name=my-agent-service \
  --app_name=my_agent \
  path/to/my_agent

Development deployment with the ADK UI:

adk deploy cloud_run \
  --project=PROJECT_ID \
  --region=REGION \
  --service_name=my-agent-service \
  --with_ui \
  path/to/my_agent

The UI is intended only for development and testing. Production deployments should normally expose the API server without --with_ui.

Important options:

• --project: Google Cloud project
• --region: Cloud Run region
• --service_name: Cloud Run service name
• --app_name: ADK API application name
• --port: Server port
• --with_ui: Include the development UI
• --trace_to_cloud: Enable Cloud Trace
• --otel_to_cloud: Enable Google Cloud observability
• --log_level: Logging level
• --adk_version: ADK version deployed
• --a2a: Enable the A2A endpoint
• --allow_origins: Configure CORS
• --trigger_sources: Enable Pub/Sub or Eventarc triggers
• --session_service_uri: Configure session storage
• --artifact_service_uri: Configure artifact storage
• --memory_service_uri: Configure memory storage

Pass additional gcloud flags after --:

adk deploy cloud_run \
  --project=PROJECT_ID \
  --region=REGION \
  path/to/my_agent \
  -- \
  --no-allow-unauthenticated \
  --min-instances=2

Official Cloud Run deployment reference

---

9. Deploy to Vertex AI Agent Engine

Using a Google Cloud project:

adk deploy agent_engine \
  --project=PROJECT_ID \
  --region=REGION \
  --display_name="My Agent" \
  path/to/my_agent

Using Express Mode:

adk deploy agent_engine \
  --api_key=API_KEY \
  path/to/my_agent

Update an existing Agent Engine resource:

adk deploy agent_engine \
  --project=PROJECT_ID \
  --region=REGION \
  --agent_engine_id=AGENT_ENGINE_ID \
  path/to/my_agent

Important options:

• --api_key: Express Mode API key
• --project: Google Cloud project
• --region: Google Cloud region
• --agent_engine_id: Update an existing deployment
• --display_name: Agent display name
• --description: Agent description
• --adk_app: Python file defining the ADK application
• --adk_app_object: Either root_agent or app
• --env_file: Environment-variable file
• --requirements_file: Python requirements file
• --agent_engine_config_file: Agent Engine configuration file
• --validate-agent-import: Validate imports before deployment
• --trace_to_cloud: Enable Cloud Trace
• --otel_to_cloud: Enable OpenTelemetry

---

10. Deploy to Google Kubernetes Engine

adk deploy gke \
  --project=PROJECT_ID \
  --region=REGION \
  --cluster_name=CLUSTER_NAME \
  path/to/my_agent

Expose it through a load balancer:

adk deploy gke \
  --project=PROJECT_ID \
  --region=REGION \
  --cluster_name=CLUSTER_NAME \
  --service_type=LoadBalancer \
  path/to/my_agent

Important options:

• --project
• --region
• --cluster_name
• --service_name
• --app_name
• --port
• --service_type=ClusterIP
• --service_type=LoadBalancer
• --with_ui
• --trace_to_cloud
• --otel_to_cloud
• --log_level
• --adk_version
• --trigger_sources
• Storage and memory service URI options

---

Evaluation and testing

11. Evaluate an agent

adk eval \
  path/to/my_agent/__init__.py \
  path/to/eval_set.json

Evaluate multiple sets:

adk eval \
  path/to/my_agent/__init__.py \
  eval_set_1.json \
  eval_set_2.json

Run selected cases:

adk eval \
  path/to/my_agent/__init__.py \
  eval_set.json:eval_1,eval_2

Print detailed results:

adk eval \
  --print_detailed_results \
  path/to/my_agent/__init__.py \
  eval_set.json

Important options:

• --config_file_path
• --print_detailed_results
• --eval_storage_uri
• --log_level
• --enable_features
• --disable_features

---

12. Display evaluation-set commands

adk eval_set --help

The eval_set group manages collections of evaluation cases.

---

13. Create an empty evaluation set

adk eval_set create \
  path/to/my_agent/__init__.py \
  EVAL_SET_ID

With Cloud Storage:

adk eval_set create \
  --eval_storage_uri=gs://BUCKET_NAME \
  path/to/my_agent/__init__.py \
  EVAL_SET_ID

---

14. Add an evaluation case

Using a scenarios file:

adk eval_set add_eval_case \
  --scenarios_file=scenarios.json \
  path/to/my_agent/__init__.py \
  EVAL_SET_ID

Using a session input file:

adk eval_set add_eval_case \
  --session_input_file=session.json \
  path/to/my_agent/__init__.py \
  EVAL_SET_ID

---

15. Generate evaluation cases

adk eval_set generate_eval_cases \
  --user_simulation_config_file=user-simulation.json \
  path/to/my_agent/__init__.py \
  EVAL_SET_ID

This uses the Vertex AI evaluation tooling to generate conversation scenarios dynamically.

---

16. Run ADK JSON tests

adk test

Specify a folder:

adk test path/to/agents

Rebuild test files by running the real agent:

adk test --rebuild path/to/agents

---

Conformance testing

17. Display conformance commands

adk conformance --help

Conformance tests check whether agent behavior remains consistent.

---

18. Record conformance tests

Non-streaming:

adk conformance record tests none

SSE streaming:

adk conformance record tests sse

Bidirectional streaming:

adk conformance record tests bidi

This feature is marked as work in progress in the current reference.

---

19. Run conformance tests

adk conformance test

Specify test folders:

adk conformance test tests/core tests/tools

Generate a Markdown report:

adk conformance test \
  --generate_report \
  --report_dir=reports

Choose replay mode:

adk conformance test \
  --mode=replay \
  tests

The documented live mode is not yet implemented in the current reference.

---

Optimization and migration

20. Optimize the root-agent instructions

adk optimize \
  --sampler_config_file_path=sampler-config.json \
  path/to/my_agent/__init__.py

With a custom optimizer configuration:

adk optimize \
  --sampler_config_file_path=sampler-config.json \
  --optimizer_config_file_path=optimizer-config.json \
  --print_detailed_results \
  path/to/my_agent/__init__.py

This uses the GEPA optimizer to improve the root agent’s instructions based on evaluation data.

---

21. Display migration commands

adk migrate --help

The migration group currently provides session-database migration.

---

22. Migrate a session database

adk migrate session \
  --source_db_url=sqlite:///old-sessions.db \
  --dest_db_url=sqlite:///new-sessions.db

Migrates session data into the latest database schema.

---

Setup commands related to ADK

These are package-management commands, not ADK subcommands.

Install Google ADK

pip install google-adk

Upgrade Google ADK

pip install --upgrade google-adk

Inspect the installed package

pip show google-adk

Confirm the executable location

which adk

Stop any running ADK server

Ctrl+C

Complete command tree

adk
├── api_server
├── conformance
│   ├── record
│   └── test
├── create
├── deploy
│   ├── agent_engine
│   ├── cloud_run
│   └── gke
├── eval
├── eval_set
│   ├── add_eval_case
│   ├── create
│   └── generate_eval_cases
├── migrate
│   └── session
├── optimize
├── run
├── test
└── web

For everyday development, the five most important commands are:

adk create my_agent
adk run my_agent
adk web .
adk api_server .
adk deploy cloud_run --project=PROJECT_ID --region=REGION my_agent

Google Cloud CLI (gcloud) – Essential Commands for Beginners

What is gcloud?

gcloud is the Google Cloud Command Line Interface (CLI). It lets you manage your Google Cloud resources directly from the Terminal instead of using the web console.

You can use gcloud to:

• Create and manage projects
• Enable APIs
• Deploy applications
• Manage virtual machines
• Work with Cloud Run
• Configure billing
• Create service accounts
• Manage AI services such as Gemini

---

Numbered gcloud Commands

1. Check whether gcloud is installed

gcloud --version

Displays the installed Google Cloud SDK version.

---

2. Update gcloud

gcloud components update

Updates the Google Cloud SDK to the latest version.

---

3. Sign in

gcloud auth login

Opens your browser so you can sign in with your Google account.

---

4. List authenticated accounts

gcloud auth list

Shows every Google account currently authenticated.

---

5. Change active account

gcloud config set account your-email@gmail.com

Sets the active Google account.

---

6. List projects

gcloud projects list

Displays every project you have access to.

---

7. Set the active project

gcloud config set project PROJECT_ID

Sets the active Google Cloud project.

Example:

gcloud config set project kiaapp-agent

All future commands use this project.

---

8. Show current project

gcloud config get-value project

Displays the currently selected project.

---

9. Describe a project

gcloud projects describe PROJECT_ID

Shows detailed information about a project.

---

10. Check billing status

gcloud billing projects describe PROJECT_ID

Shows whether billing is enabled for the project.

---

11. List billing accounts

gcloud billing accounts list

Displays available billing accounts.

---

12. Enable an API

gcloud services enable API_NAME

Enables a Google Cloud service.

Example:

gcloud services enable aiplatform.googleapis.com

Enables the Vertex AI API.

---

13. List enabled APIs

gcloud services list

Shows APIs enabled in the current project.

---

14. List all available APIs

gcloud services list --available

Lists every Google Cloud API available for use.

---

15. List configurations

gcloud config configurations list

Shows all saved gcloud configuration profiles.

---

16. Show current configuration

gcloud config list

Displays the active configuration settings.

---

17. Create a configuration

gcloud config configurations create my-config

Creates a new configuration profile.

---

18. Activate a configuration

gcloud config configurations activate my-config

Switches to another configuration.

---

19. Deploy an application

gcloud run deploy

Deploys your application to Cloud Run.

---

20. List Cloud Run services

gcloud run services list

Displays deployed Cloud Run services.

---

21. Delete a service

gcloud run services delete SERVICE_NAME

Deletes a Cloud Run service.

---

22. List virtual machines

gcloud compute instances list

Displays Compute Engine virtual machine instances.

---

23. Create a virtual machine

gcloud compute instances create VM_NAME

Creates a new Compute Engine virtual machine.

---

24. List Cloud Storage buckets

gcloud storage buckets list

Displays Cloud Storage buckets.

---

25. Create a Cloud Storage bucket

gcloud storage buckets create gs://BUCKET_NAME

Creates a new Cloud Storage bucket.

---

26. List service accounts

gcloud iam service-accounts list

Displays all service accounts in the project.

---

27. Create a service account

gcloud iam service-accounts create NAME

Creates a new service account.

---

28. Enable the Vertex AI API

gcloud services enable aiplatform.googleapis.com

Enables Vertex AI so you can use AI models such as Gemini.

---

29. Check the active project before using Gemini

gcloud config get-value project

Confirms that Gemini requests will use the correct Google Cloud project.

---

30. Show general help

gcloud help

Displays general help for the gcloud command-line tool.

---

31. Show Compute Engine help

gcloud compute --help

Displays all Compute Engine commands and their usage.

---

32. Show Cloud Run help

gcloud run --help

Displays all Cloud Run commands and their usage.

---

33. Print current SDK information

gcloud info

Displays Google Cloud SDK installation details, environment information, and the active configuration.

Before comparing every AI model, one important point needs to be clear: ChatGPT is not the model. ChatGPT is the app. GPT is the model behind it. Think of ChatGPT as a door you walk through to access GPT. Copilot, Gemini, and Claude work in a similar way. Each has a different interface, logo, and product experience, but behind them sits a powerful AI model doing the real work.

Most AI models are trained on huge amounts of text, code, books, articles, and websites. They do not “memorize” information in the way humans do. Instead, they learn patterns. At the most basic level, many of these models predict the next word, or more accurately, the next token. That may sound simple, but when done extremely well, it can produce essays, explain science, write code, summarize documents, and answer complex questions. It is like autocomplete, but trained on a massive amount of information and improved to understand context far better.

Two important ideas help explain why some models feel smarter than others: parameters and context window. Parameters are like the internal settings the model uses to recognize patterns. More parameters usually mean the model can understand more complex relationships. The context window is how much information the model can keep in mind during a conversation. A larger context window means you can give it longer documents, bigger code files, or extended conversations without it losing track.

Some newer models also include stronger reasoning abilities. These models do not just answer immediately; they take more time to work through the problem. That makes them slower, but often much better for math, logic, coding, planning, and multi-step tasks.

GPT, developed by OpenAI, is one of the most well-known model families because it is strong across many tasks. It can help with writing, analysis, coding, image-related work, voice interaction, and general problem solving. Its biggest advantage is not only the model itself, but the ecosystem around it. Many apps, plugins, businesses, and third-party tools are built around OpenAI’s GPT models, which makes them widely useful in everyday work.

Gemini, developed by Google DeepMind, is especially powerful because of its integration with Google’s ecosystem. If someone already uses Gmail, Google Docs, Sheets, Android, Search, or Maps, Gemini fits naturally into that workflow. Its strength is not only answering questions, but also helping inside the tools people already use every day.

Claude, developed by Anthropic, is often preferred for coding, long-document analysis, and deep reasoning. Many developers recommend Claude because it can review code, explain complex systems, and produce well-structured summaries. It is also known for providing more balanced and direct feedback rather than simply agreeing with every suggestion.

Grok, developed by xAI, is best known for its integration with X (formerly Twitter) and its ability to analyze real-time conversations and trending topics. This makes it particularly useful for following breaking news, monitoring public sentiment, and understanding online discussions as they happen.

DeepSeek, developed by DeepSeek AI, represents another important direction in AI development. As an open-source model, it allows users to download and run AI locally instead of relying on cloud services. Other open-source models such as Llama from Meta, Qwen from Alibaba, and Mistral from Mistral AI also give users greater control, improved privacy, and the ability to run powerful AI models on their own hardware.

Beyond conversational models, there are also specialized AI systems. Midjourney is widely regarded for producing highly artistic images. DALL·E, developed by OpenAI, is easy to use and performs particularly well when generating images containing readable text. Flux and Stable Diffusion offer greater customization and are popular among users who want to generate images locally.

For AI-generated video, Sora from OpenAI, Runway, and Kling are among the leading platforms, while Suno and Udio have become popular for creating complete songs with vocals and instruments from simple text prompts.

The next major evolution is AI agents. Instead of simply answering questions, AI agents can browse websites, execute code, manage files, fill out forms, and complete complex multi-step tasks with limited human supervision. Systems such as OpenAI Operator, Google Project Mariner, and Anthropic Computer Use demonstrate how AI is evolving from a conversational assistant into an autonomous digital worker.

The best strategy today is not to rely on a single AI model. Use GPT for general-purpose work, Gemini if you live inside Google’s ecosystem, Claude for programming and detailed analysis, Perplexity for research with source citations, and Llama or DeepSeek if privacy and local execution are priorities. AI is becoming less like one universal application and more like a complete toolbox, where choosing the right model for the right task produces the best results.

If you want to build your first AI agent on Windows safely without affecting your main Python installation, this is the cleanest way.

---

1. Create your project folder(Command Prompt / PowerShell)

This creates a new workspace for your AI project.

mkdir ai-agent

---

2. Move into your project folder

Enter your project directory.

cd ai-agent

Think of this as entering your AI workshop.

---

3. Create a virtual environment(Important for safety)

This creates a private Python environment for your project.

It protects your main Python installation from package conflicts.

python -m venv venv

This creates:

ai-agent\
└── venv\

Why this matters:

Without it, installing packages can break:

• global Python packages
• other projects
• dependency versions

With it, your project stays isolated.

---

4. Activate the virtual environment

On Windows:

venv\Scripts\activate

After activation you should see:

(venv) C:\Users\YourName\ai-agent>

That means your isolated environment is active.

---

5. Install Google ADK

Install the toolkit:

pip install google-adk

This installs only inside your venv.

---

6. Create your Python file

Create your agent file:

type nul > agent.py

Or create it manually inside your editor.

---

7. Open your project in Cursor / VS Code

Open the whole:

ai-agent

folder.

Then open:

agent.py

Paste:

from google.adk.agents import Agent

root_agent = Agent(
    name="hello_agent",
    model="gemini-2.0-flash",
    instruction="You are a helpful AI assistant."
)

Save it.

---

8. Add your Google API key

Connect your agent to Google Gemini:

set GOOGLE_API_KEY=your_key_here

Replace with your real API key.

Important:

This only works for the current terminal session.

---

9. Start your AI agent

Run:

adk web

This launches the local web interface.

---

10. Open in browser

Go to:

http://localhost:8000

Your AI agent is now live.

You can chat with it directly.

---

What you built

You now have:

✅ Your first AI agent
✅ Running locally on Windows
✅ Connected to Google Gemini
✅ Protected inside an isolated Python environment
✅ Ready to expand with APIs, tools, memory, and workflows

Think of it like:

• Project folder = workshop
• venv = protected lab
• agent.py = brain
• API key = power source
• adk web = engine
• localhost = control center

Important safety rules

🚫 Never share your API key
🚫 Never install packages globally unless needed
🚫 Never skip the virtual environment
🚫 Avoid pip install outside venv