Google ADKgoogle-adk 2.8 · Python 3.10+
1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
28 small wins to finish your pathNext lesson

The dev UI, the CLI and a server

Everything so far ran inside a Python script. ADK ships three other ways to run the same agent, and each one is a single command.

bash
adk --help
Captured from a real run
Usage: adk [OPTIONS] COMMAND [ARGS]...

  Agent Development Kit CLI tools.

Commands:
  api_server   Starts a FastAPI server for agents.
  create       Creates a new app in the current folder with prepopulated...
  deploy       Deploys agent to hosted environments.
  eval         Evaluates an agent given the eval sets.
  run          Runs an agent.
  web          Starts a FastAPI server with Web UI for agents.

That is the real output, trimmed to the commands worth knowing on day one. All of them need the layout from lesson 1: a folder with an agent.py that has a variable called root_agent.

The three you will use

CommandWhat it gives you
adk run my_agentA conversation in your terminal
adk webA browser interface, with the events of each run laid out
adk api_serverAn HTTP server, so something else can call your agent

The web one is the one to try first. It shows the same events you have been printing since lesson 6, but as a tree you can click through, which makes a multi-agent run much easier to follow than a wall of text.

Why the server matters

api_server is what turns an agent into something a product can use. It is the same agent, the same sessions and the same state, reachable over HTTP instead of from a Python file. Nothing about your code changes.

text
my_agent/
    agent.py      # root_agent lives here
    .env          # keys, for a real model
    __init__.py

The other commands are for later: eval runs test sets against an agent, and deploy pushes it to Cloud Run, GKE or the managed runtime. Both have their own sections in the documentation, and neither changes how the agent is written.

Worth doing once
Point adk web at the agent you built in part 4 and open the event view. Watching a transfer happen in a tree is the fastest way to understand what the parts of this course actually do together.
Try it yourself
  • Put one of your agents in an agent.py and run adk run against it.
  • Start adk web and click through the events of a sequential agent.

This is what real progress feels like.