1. Introduzione
Panoramica
In questo codelab creerai un assistente AI personale che ti aiuterà ad analizzare i dati aziendali ed eseguire altre attività tramite un'interfaccia utente di chat. Utilizzerai un servizio Cloud Run per ospitare il tuo agente personale.
L'agente utilizzerà le sandbox di Cloud Run. Le sandbox di Cloud Run sono un ambiente di runtime nativo, sicuro e velocissimo creato appositamente per l'esecuzione di codice non attendibile e carichi di lavoro degli agenti, a partire da millisecondi. La sandbox consente all'agente AI di scrivere, eseguire e testare dinamicamente il codice al volo per risolvere problemi analitici complessi.
Nota: per garantire un'esperienza di sviluppo senza interruzioni durante l'esecuzione in locale rispetto alla produzione:
- In produzione (sandbox di Cloud Run): l'agente esegue il codice in modo sicuro all'interno di un ambiente di test containerizzato isolato tramite un file binario sandbox dedicato (/usr/local/gcp/bin/sandbox).
- In locale (sul tuo computer): quando viene eseguita in locale, l'app rileva che l'ambiente sandbox di produzione non è presente (IS_LOCAL_MODE = True). L'agente esegue script Python e comandi della shell direttamente sul terminale di sistema della macchina host locale.
Cosa creerai:
In questo scenario, gestisci una caffetteria in una città universitaria che si sta preparando per un fine settimana di laurea. Devi fare in modo che l'agente faccia un controllo incrociato ai dati grezzi del punto vendita (POS) con il programma delle cerimonie dell'università per scoprire colli di bottiglia operativi nascosti.
L'agente utilizza una sandbox sicura per scrivere ed eseguire script Python, analizzando la complessità delle bevande rispetto al numero di cassieri per consigliare modifiche al personale e all'inventario.
L'agente invia una notifica al proprietario tramite un'interfaccia utente di chat simulata con i consigli mirati per l'inventario e il personale. Attende l'autorizzazione esplicita prima di aggiornare il foglio di lavoro con le attività operative da svolgere per il responsabile della caffetteria.
Obiettivi didattici
- Come creare un servizio Cloud Run
- Come eseguire il deployment di un agente ADK su un servizio Cloud Run
- Come fare in modo che un agente esegua il codice in una sandbox all'interno di un servizio Cloud Run
- Come creare un'interfaccia utente di chat utilizzando WebSocket per interagire con l'agente in background
2. Configurazione e requisiti
Configura l'ID progetto e la regione.
GOOGLE_CLOUD_PROJECT=<YOUR_PROJECT_ID>
REGION=us-west2
gcloud config set project $GOOGLE_CLOUD_PROJECT
gcloud config set run/region $REGION
Di seguito sono riportate le variabili di ambiente che verranno utilizzate in questo codelab. Puoi salvarle in un file di ambiente e "originarlo". Assicurati di impostare correttamente il valore dell'ID progetto e, facoltativamente, della regione.
SA_NAME=coffee-shop-agent-sa
SERVICE_ACCOUNT_ADDRESS=$SA_NAME@$GOOGLE_CLOUD_PROJECT.iam.gserviceaccount.com
Abilita le API necessarie per questo codelab
gcloud services enable --project $GOOGLE_CLOUD_PROJECT \
run.googleapis.com \
cloudbuild.googleapis.com \
artifactregistry.googleapis.com \
sheets.googleapis.com \
aiplatform.googleapis.com
3. Crea un account di servizio e un foglio di lavoro
È consigliabile creare un account di servizio dedicato per il servizio Cloud Run per concedere solo i ruoli richiesti necessari.
Crea un account di servizio per il servizio Cloud Run
gcloud iam service-accounts create $SA_NAME \
--description="Service account for the Coffee Shop Agent Codelab" \
--display-name="Coffee Shop Agent SA"
Poiché l'agente utilizza le API Gemini (GOOGLE_GENAI_USE_VERTEXAI=1), concedi a questo account di servizio il ruolo Utente della piattaforma agente nel tuo progetto.
gcloud projects add-iam-policy-binding $GOOGLE_CLOUD_PROJECT \
--member="serviceAccount:$SERVICE_ACCOUNT_ADDRESS" \
--role="roles/aiplatform.user"
Per consentirti di eseguire e testare l'agente in locale utilizzando questo account di servizio, concedi alla tua identità Google Cloud personale l'autorizzazione a rappresentare questo account di servizio:
gcloud iam service-accounts add-iam-policy-binding \
coffee-shop-agent-sa@$GOOGLE_CLOUD_PROJECT.iam.gserviceaccount.com \
--member="user:$(gcloud config get-value account)" \
--role="roles/iam.serviceAccountTokenCreator"
Crea il foglio di lavoro
Questo foglio di lavoro rappresenta le vendite avvenute l'anno scorso durante il fine settimana di laurea.
- Crea un nuovo Foglio Google in Google Drive.
- Copia i seguenti valori separati da virgola (CSV) nel nuovo Foglio Google (ad es. seleziona la cella A1 e incolla).
Day,Time,Drip_Coffee,Cold_Brew,Extra_Espresso,Alt_Milk_Oz,Pastries,Cashiers_Working,Wait_Time_Minutes
Saturday,07:00:00,95,15,5,30,80,2,9
Saturday,08:00:00,80,25,10,35,60,2,7
Saturday,09:00:00,30,30,20,40,20,1,4
Saturday,10:00:00,40,130,95,50,25,2,4
Saturday,11:00:00,25,45,25,40,15,1,5
Saturday,12:00:00,30,35,15,45,20,1,3
Saturday,13:00:00,15,20,10,30,10,1,2
Saturday,14:00:00,45,60,30,65,30,2,8
Saturday,15:00:00,20,30,15,35,15,1,3
Saturday,16:00:00,25,35,10,50,20,1,4
Saturday,17:00:00,10,20,5,35,10,1,2
Saturday,18:00:00,20,40,10,190,65,2,12
Saturday,19:00:00,10,15,5,40,15,1,2
Sunday,07:00:00,90,10,5,35,75,2,8
Sunday,08:00:00,75,20,10,40,65,2,6
Sunday,09:00:00,25,25,15,35,15,1,3
Sunday,10:00:00,60,35,15,60,50,2,7
Sunday,11:00:00,20,30,10,35,10,1,3
Sunday,12:00:00,25,40,10,55,20,1,3
Sunday,13:00:00,15,25,5,40,10,1,2
Sunday,14:00:00,30,50,15,180,60,2,11
Sunday,15:00:00,15,25,10,45,15,1,3
Sunday,16:00:00,20,45,20,40,15,1,4
Sunday,17:00:00,10,25,10,30,10,1,2
Sunday,18:00:00,25,145,110,45,20,2,16
Sunday,19:00:00,15,30,15,35,10,1,4
- Con i dati ancora selezionati, fai clic su Dati > Suddividi il testo in colonne nel menu in alto di Fogli Google.
Assicurati che l'account di servizio abbia accesso come Editor al foglio di lavoro. È simile a come daresti l'accesso a un collega.
- Copia l'indirizzo email completo del nuovo service account
echo $SERVICE_ACCOUNT_ADDRESS
- Da Foglio Google, fai clic su Condividi nell'angolo in alto a destra
- Incolla l'indirizzo email dell'account di servizio, imposta l'autorizzazione su Editor e fai clic su Condividi. (Puoi deselezionare Invia notifica).
- Registra l'ID del foglio di lavoro che passerai all'agente, ad es. https://docs.google.com/spreadsheets/d/<YOUR_SPREADSHEET_ID>/edit?gid=0#gid=0
SPREADSHEET_ID=<THE SPREADHSEET ID FROM ITS URL>
4. Crea l'agente ADK
Innanzitutto, crea una directory per il codice.
mkdir coffee-mgr-agent && cd coffee-mgr-agent
Crea un file requirements.txt.
fastapi>=0.100.0
uvicorn>=0.22.0
google-adk>=1.27.1
google-auth
google-api-python-client
Crea un Dockerfile
FROM python:3.11-slim
ENV PYTHONUNBUFFERED=1
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY main.py .
EXPOSE 8080
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8080"]
Crea un file main.py
import os
import asyncio
import subprocess
from pathlib import Path
from typing import List
from fastapi import FastAPI, HTTPException, WebSocket, WebSocketDisconnect
from fastapi.responses import HTMLResponse
from pydantic import BaseModel
from google.adk.agents import LlmAgent as Agent
from google.adk.tools import FunctionTool
from google.adk.apps import App
from google.adk import Runner
from google.adk.sessions import InMemorySessionService
from google.genai import types
SANDBOX_CLI = '/usr/local/gcp/bin/sandbox'
IS_LOCAL_MODE = not Path(SANDBOX_CLI).exists()
active_connections: list[WebSocket] = []
SPREADSHEET_ID = os.environ.get("SPREADSHEET_ID")
# And
def run_sandbox_process(args: list[str]):
cmd = args[2:] if IS_LOCAL_MODE and args[:2] == ['do', '--'] else ([SANDBOX_CLI] + args if not IS_LOCAL_MODE else args)
return subprocess.run(cmd, capture_output=True, text=True, timeout=10)
def execute_sandbox_command(command: str) -> str:
"""Executes arbitrary POSIX shell/bash commands inside sandbox."""
mode = "LOCAL" if IS_LOCAL_MODE else "CLOUD RUN SANDBOX"
print(f"[ADK Sandbox Tool] Starting {mode} shell run...")
try:
res = run_sandbox_process(['do', '--', '/bin/sh', '-c', command])
if res.returncode != 0:
return f"Execution Failed!\nExit Code: {res.returncode}\nStdout:\n{res.stdout}\nStderr:\n{res.stderr}"
return res.stdout
except Exception as err:
return f"Internal Sandbox Tool Error: {str(err)}"
def get_sheets_service():
"""Initializes and returns the Google Sheets client service."""
from google.auth import default
from googleapiclient.discovery import build
credentials, _ = default(scopes=[
'https://www.googleapis.com/auth/spreadsheets',
'https://www.googleapis.com/auth/cloud-platform'
])
return build('sheets', 'v4', credentials=credentials)
def read_spreadsheet_values(spreadsheet_id: str, range_name: str) -> str:
"""Reads a range of cells from a Google Spreadsheet."""
try:
service = get_sheets_service()
result = service.spreadsheets().values().get(spreadsheetId=spreadsheet_id, range=range_name).execute()
rows = result.get('values', [])
return str(rows) if rows else "No data found in the specified range."
except Exception as e:
return f"Read Error: {str(e)}"
def update_spreadsheet_values(spreadsheet_id: str, range_name: str, values: List[List[str]]) -> str:
"""Updates a range of cells in a Google Spreadsheet with the provided values."""
try:
service = get_sheets_service()
result = service.spreadsheets().values().update(
spreadsheetId=spreadsheet_id, range=range_name,
valueInputOption="USER_ENTERED", body={'values': values}).execute()
return f"Successfully updated {result.get('updatedCells')} cells in {range_name}."
except Exception as e:
return f"Write Error: {str(e)}"
def create_spreadsheet_tab(spreadsheet_id: str, tab_name: str) -> str:
"""Creates a new sheet tab in a Google Spreadsheet if it doesn't already exist."""
try:
service = get_sheets_service()
# Check if tab exists
spreadsheet = service.spreadsheets().get(spreadsheetId=spreadsheet_id).execute()
for sheet in spreadsheet.get('sheets', []):
if sheet.get('properties', {}).get('title') == tab_name:
return f"Sheet tab '{tab_name}' already exists."
# Create tab
body = {'requests': [{'addSheet': {'properties': {'title': tab_name}}}]}
service.spreadsheets().batchUpdate(spreadsheetId=spreadsheet_id, body=body).execute()
return f"Successfully created new sheet tab '{tab_name}'."
except Exception as e:
return f"Error creating sheet tab: {str(e)}"
# ==========================================
# 3. ADK AGENT & RUNNER SETUP
# ==========================================
root_agent = Agent(
name='secure_coding_assistant',
description='ADK agent capable of executing shell commands and managing Google Spreadsheets.',
model=os.environ.get('GEMINI_MODEL', 'gemini-3.1-flash-lite'),
instruction=(
f'You are an expert AI Business Analyst for a coffee shop during university graduation weekend.\n'
f'The Google Spreadsheet ID you are managing is: "{SPREADSHEET_ID}". Use this ID for all sheet operations.\n'
'1. Comparative Analysis Policy:\n'
f' - Ingest historical POS data from the "POS-2025" sheet tab using read_spreadsheet_values with spreadsheet_id="{SPREADSHEET_ID}".\n'
' - Receive the current graduation schedule directly from the manager\'s prompt (the manager will paste it and indicate it is the same schedule sequence as last year).\n'
' - Write a python3 script via the sandbox tool to:\n'
' a. Correlate the 2025 product spikes (Cold Brew, Alt Milk, Extra Espresso) with the specific ceremonies ending at those times.\n'
' b. Map those beverage profiles to the pasted schedule (which is the same sequence) to predict exactly when and where the 2026 spikes will occur.\n'
' c. Identify expected wait-time bottlenecks in 2026 based on the 2025 wait times for those same profiles.\n'
'2. Bottleneck Diagnostics (Playbook):\n'
' - If a predicted high-volume slot in 2026 is expected to have Wait_Time_Minutes > 10:\n'
' - If Cashiers_Working < 2: Recommend scheduling another cashier.\n'
' - If Cashiers_Working == 2 and complex items (Cold Brew, Extra Espresso, Alt Milk) spike: Deduce that the bottleneck is barista output, not cashiers. Recommend adding a "Support Barista" role to handle fulfillment.\n'
'3. Human-in-the-Loop Policy:\n'
' - Present your detailed data discoveries, wait-time bottlenecks, and actionable recommendations (stocking and staffing changes) to the manager.\n'
' - Highlight only two or three findings for specific ceremonies.\n'
' - Frame your recommendations as a clean list of suggested tasks for the manager\'s TODO list.\n'
' - Explicitly ask: "Would you like me to add these tasks to your \'TODO-2026\' TODO list?"\n'
' - Do NOT modify any spreadsheet data until explicit approval is given.\n'
'4. Post-Approval Policy:\n'
f' - Upon receiving explicit user approval, first verify if the "TODO-2026" sheet tab exists in spreadsheet "{SPREADSHEET_ID}".\n'
f' - If the "TODO-2026" sheet tab does not exist, use the tool create_spreadsheet_tab to create it in spreadsheet "{SPREADSHEET_ID}".\n'
f' - Once the tab exists, use update_spreadsheet_values to append the approved adjustments as tasks to the "TODO-2026" sheet tab.\n'
' - Write the rows under the headers: Task (the actionable job, e.g., "Schedule a Support Barista role for Saturday morning"), Category ("Staffing" or "Inventory"), Ceremony, and Date_Added (today\'s date).\n'
' - Always confirm to the user exactly what tasks you have written to their "TODO-2026" TODO list.'
),
tools=[
FunctionTool(func=execute_sandbox_command),
FunctionTool(func=read_spreadsheet_values),
FunctionTool(func=update_spreadsheet_values),
FunctionTool(func=create_spreadsheet_tab)
]
)
adk_app = App(name="secure_sandbox_app", root_agent=root_agent)
runner = Runner(app=adk_app, session_service=InMemorySessionService(), auto_create_session=True)
app = FastAPI(title="Secure ADK Sandbox Assistant")
# ==========================================
# 4. ENDPOINTS & WEBSOCKET ROUTING
# ==========================================
@app.websocket("/ws")
async def websocket_endpoint(websocket: WebSocket):
await websocket.accept()
active_connections.append(websocket)
await websocket.send_text("🔌 System: Connected. Agent is ready...")
try:
while True:
owner_reply = await websocket.receive_text()
print(f"Owner replied via WS: {owner_reply}")
await websocket.send_text("_Agent is running tools and thinking..._")
new_message = types.Content(parts=[types.Part(text=owner_reply)])
events = await asyncio.to_thread(
runner.run,
user_id="local_user",
session_id="local_session",
new_message=new_message
)
final_response = "".join(
part.text
for event in events
if event.content and event.content.parts
for part in event.content.parts
if part.text
) or "Agent completed execution updates without text output."
await websocket.send_text(final_response.strip())
except WebSocketDisconnect:
active_connections.remove(websocket)
class UserPrompt(BaseModel):
prompt: str
@app.post("/chat")
def chat_with_agent(payload: UserPrompt):
"""Fallback HTTP POST endpoint if UI is not used."""
try:
events = runner.run(
user_id="local_user",
session_id="local_session",
new_message=types.Content(parts=[types.Part(text=payload.prompt)])
)
final_response = "".join(
part.text
for event in events
if event.content and event.content.parts
for part in event.content.parts
if part.text
)
return {"status": "success", "response": final_response.strip()}
except Exception as e:
raise HTTPException(status_code=500, detail=f"Agent loop failed: {str(e)}")
@app.get("/", response_class=HTMLResponse)
async def get_chat_ui():
"""Serves the warm coffee-themed chat interface."""
return """
<!DOCTYPE html>
<html>
<head>
<title>Coffee Shop Agent</title>
<!-- Load marked.js for client-side markdown rendering -->
<script src="https://cdn.jsdelivr.net/npm/marked/marked.min.js"></script>
<style>
body { display: flex; height: 100vh; margin: 0; font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; }
#sidebar { width: 250px; background: #3E2723; color: #EFEBE9; padding: 20px; }
#sidebar h2 { font-size: 1.3em; margin-top: 10px; color: #FFF; border-bottom: 1px solid #5D4037; padding-bottom: 10px; }
#main { flex-grow: 1; display: flex; flex-direction: column; background: #FAF8F6; }
#chat-history { flex-grow: 1; padding: 20px; overflow-y: auto; background: #F5EFEB; }
#input-area { padding: 20px; border-top: 1px solid #D7CCC8; background: #FAF8F6; display: flex;}
input {
flex-grow: 1;
padding: 12px;
border-radius: 6px;
border: 1px solid #D7CCC8;
margin-right: 10px;
font-size: 1em;
background: #FFF;
transition: border-color 0.2s, box-shadow 0.2s;
}
input:focus {
outline: none;
border-color: #8D6E63;
box-shadow: 0 0 0 2px rgba(141, 110, 99, 0.25);
}
button {
padding: 10px 24px;
background: #6D4C41;
color: white;
border: none;
border-radius: 6px;
cursor: pointer;
font-weight: bold;
font-size: 1em;
transition: background-color 0.2s, transform 0.1s;
}
button:hover {
background: #5D4037;
}
button:active {
background: #4E342E;
transform: scale(0.98);
}
.message { margin-bottom: 15px; padding: 12px 16px; border-radius: 8px; max-width: 85%; line-height: 1.5; }
.user-msg { background: #EFEBE9; color: #3E2723; align-self: flex-end; margin-left: auto; border: 1px solid #D7CCC8;}
.agent-msg { background: #fff; color: #3E2723; border: 1px solid #E0DCD8; box-shadow: 0 1px 3px rgba(62,39,35,0.06); }
.msg-header { display: flex; justify-content: space-between; align-items: center; margin-bottom: 6px; }
.agent-name { font-weight: bold; color: #5D4037; margin: 0; font-size: 0.95em;}
.user-name { font-weight: bold; color: #8D6E63; margin: 0; font-size: 0.95em;}
.msg-timestamp { font-size: 0.8em; color: #8D6E63; font-weight: normal; }
.day-divider {
display: flex;
align-items: center;
text-align: center;
color: #8D6E63;
margin: 20px 0;
font-size: 0.85em;
font-weight: bold;
}
.day-divider::before, .day-divider::after {
content: '';
flex: 1;
border-bottom: 1px solid #D7CCC8;
}
.day-divider:not(:empty)::before {
margin-right: .5em;
}
.day-divider:not(:empty)::after {
margin-left: .5em;
}
/* Markdown Styling inside Messages */
.message p { margin: 4px 0 8px 0; }
.message p:last-child { margin-bottom: 0; }
.message ul, .message ol { margin: 4px 0 8px 0; padding-left: 20px; }
.message li { margin-bottom: 3px; }
.message h1, .message h2, .message h3, .message h4 { margin: 12px 0 6px 0; font-size: 1.15em; color: #3E2723; font-weight: bold; }
.message h1:first-child, .message h2:first-child, .message h3:first-child { margin-top: 0; }
/* Table Styling */
.message table { border-collapse: collapse; width: 100%; margin: 10px 0; font-size: 0.95em; }
.message th, .message td { border: 1px solid #D7CCC8; padding: 8px 10px; text-align: left; }
.message th { background-color: #F5EFEB; font-weight: bold; color: #3E2723; }
.message tr:nth-child(even) { background-color: #FAF8F6; }
/* Code / Blockquote styling */
.message code { background: #EFEBE9; padding: 2px 4px; border-radius: 3px; font-family: monospace; font-size: 0.9em; color: #5D4037; }
.message pre { background: #F5EFEB; padding: 10px; border-radius: 5px; overflow-x: auto; margin: 8px 0; border: 1px solid #D7CCC8; }
.message pre code { background: none; padding: 0; }
.message blockquote { margin: 8px 0; padding-left: 12px; border-left: 4px solid #6D4C41; color: #5D4037; }
</style>
</head>
<body>
<div id="sidebar">
<h2>☕ Coffee Shop Monitor</h2>
<p>Monitoring sheet...</p>
</div>
<div id="main">
<div id="chat-history"></div>
<div id="input-area">
<input type="text" id="msg" placeholder="Message Coffee Shop Monitor..." onkeypress="if(event.key === 'Enter') sendMessage()">
<button onclick="sendMessage()">Send</button>
</div>
</div>
<script>
const protocol = window.location.protocol === 'https:' ? 'wss:' : 'ws:';
const ws = new WebSocket(`${protocol}//${window.location.host}/ws`);
const history = document.getElementById('chat-history');
// Scroll to the bottom on load to show latest messages
history.scrollTop = history.scrollHeight;
ws.onmessage = function(event) {
// Parse markdown content from the agent
const parsedHtml = marked.parse(event.data);
const timeStr = new Date().toLocaleTimeString([], { hour: '2-digit', minute: '2-digit' });
history.innerHTML += `
<div class="message agent-msg">
<div class="msg-header">
<span class="agent-name">Inventory Agent APP</span>
<span class="msg-timestamp">${timeStr}</span>
</div>
<div>${parsedHtml}</div>
</div>`;
history.scrollTop = history.scrollHeight;
};
function sendMessage() {
const input = document.getElementById('msg');
const text = input.value;
if (!text) return;
// Parse user markdown if any (optional, but keeps styling consistent)
const parsedHtml = marked.parse(text);
const timeStr = new Date().toLocaleTimeString([], { hour: '2-digit', minute: '2-digit' });
history.innerHTML += `
<div class="message user-msg">
<div class="msg-header">
<span class="user-name">You</span>
<span class="msg-timestamp">${timeStr}</span>
</div>
<div>${parsedHtml}</div>
</div>`;
ws.send(text);
input.value = '';
history.scrollTop = history.scrollHeight;
}
</script>
</body>
</html>
"""
if __name__ == "__main__":
import uvicorn
port_val = int(os.environ.get('PORT', 8080))
uvicorn.run(app, host='0.0.0.0', port=port_val)
5. Esegui il deployment del servizio Cloud Run
Anziché creare un'immagine, utilizzerai i buildpack di Google Cloud per trasformare il codice sorgente dell'applicazione in un'immagine container pronta per la produzione per Cloud Run.
gcloud beta run deploy coffee-mgr-agent \
--source=. \
--region=$REGION \
--sandbox-launcher \
--max-instances=1 \
--session-affinity \
--allow-unauthenticated \
--no-cpu-throttling \
--set-env-vars GOOGLE_GENAI_USE_VERTEXAI=1,GOOGLE_CLOUD_PROJECT=$GOOGLE_CLOUD_PROJECT,GOOGLE_CLOUD_LOCATION=global,SPREADSHEET_ID=$SPREADSHEET_ID \
--service-account $SERVICE_ACCOUNT_ADDRESS
6. Chatta con il tuo agente
Dopo il deployment, vedrai URL: https://coffee-mgr-agent-YOUR_PROJECT_ID.YOUR_REGION.run.app
Apri questo URL nel browser.
Invia il seguente prompt all'agente.
The 2026 graduation schedule was just posted. It's the same schedule as last year. Can you review last year's POS data to help me prepare for this year?
Saturday, June 13:
College of Business (8:30 a.m.)
College of Science and Mathematics (12:30 p.m.)
College of Liberal Arts (4:30 p.m.)
Sunday, June 14:
College of Agriculture (8:30 a.m.)
College of Architecture (12:30 p.m.)
College of Engineering (4:30 p.m.)
L'agente dovrebbe rispondere con consigli simili ai seguenti:
Based on the 2025 POS data, I have correlated graduation ceremony patterns with beverage demand spikes. Last year, the post-ceremony windows consistently experienced high demand for complex items (Cold Brew, Extra Espresso, and Alt Milk). Key Data Discoveries for 2026 - College of Business (Saturday 8:30 a.m. ceremony): - Significant demand spike observed at 10:00 a.m. Saturday. - Data: 130 Cold Brews and 95 Extra Espresso shots. This is a primary bottleneck for barista output. - College of Engineering (Sunday 4:30 p.m. ceremony): - Significant demand spike observed at 6:00 p.m. Sunday. - Data: 145 Cold Brews and 110 Extra Espresso shots. This is the highest intensity period of the weekend. - Alt-Milk Surge (Saturday/Sunday 2:00 p.m. post-ceremonies): - Consistent, high-volume Alt Milk consumption (65-180 oz) after the 12:30 p.m. ceremonies. Diagnostic & Staffing Recommendations Given that these complex beverage items (Cold Brew, Extra Espresso) correlate directly with your high-traffic post-ceremony windows, the risk of hitting wait times > 10 minutes is high. Assuming you typically have 2 cashiers, my diagnostic is that these are barista output bottlenecks, not cashier bottlenecks. Suggested Tasks for your TODO-2026 List: 1. Staffing: Schedule a dedicated "Support Barista" for Saturday 9:30 a.m. – 11:00 a.m. to handle the Cold Brew/Espresso rush. 2. Staffing: Schedule a dedicated "Support Barista" for Sunday 5:30 p.m. – 7:00 p.m. to manage the high volume after the Engineering ceremony. 3. Inventory: Pre-stage extra Alt Milk cartons at the mobile station for Saturday 1:30 p.m. and Sunday 1:30 p.m. shifts. Would you like me to add these tasks to your 'TODO-2026' TODO list?
L'agente ti chiederà (Human-in-the-Loop) l'autorizzazione per creare/aggiornare il foglio della scheda TODO-2026.
Rispondi all'agente con Yes.
Ora, quando guardi il foglio della scheda TODO-2026, vedrai qualcosa di simile a quanto segue:
Task Category Ceremony Date_Added Schedule a Support Barista role for Saturday morning rush Staffing College of Business 2025-05-22 Schedule a Support Barista role for Sunday evening rush Staffing College of Engineering 2025-05-22 Pre-stage extra Alt Milk cartons at mobile station Inventory College of Science/Math & Architecture 2025-05-22
7. Complimenti!
Complimenti per aver completato il codelab.
Ti consigliamo di consultare la documentazione di Cloud Run, la documentazione delle sandbox di Cloud Run e la documentazione di ADK.
Argomenti trattati
- Come creare un servizio Cloud Run
- Come eseguire il deployment di un agente ADK su un servizio Cloud Run
- Come fare in modo che un agente esegua il codice in una sandbox all'interno di un servizio Cloud Run
8. Libera spazio
Per evitare che al tuo account Google Cloud vengano addebitati costi relativi alle risorse utilizzate in questo tutorial, puoi eliminare il progetto o le singole risorse.
Opzione 1: elimina le risorse
Di seguito è riportata una breve panoramica dei passaggi per liberare spazio nel progetto ed evitare addebiti non necessari.
Elimina il servizio Cloud Run
gcloud run services delete coffee-mgr-agent --region $REGION
Elimina l'account di servizio: rimuovi l'account di servizio del foglio di lavoro dedicato
gcloud iam service-accounts delete $SERVICE_ACCOUNT_ADDRESS
Opzione 2: elimina il progetto
Per eliminare l'intero progetto, vai a Gestisci risorse, seleziona il progetto creato nel passaggio 2 e scegli Elimina. Se elimini il progetto, dovrai cambiare progetto in Cloud SDK. Puoi visualizzare l'elenco di tutti i progetti disponibili eseguendo gcloud projects list. Se preferisci utilizzare la riga di comando, puoi utilizzare anche questo comando:
gcloud projects delete ${GOOGLE_CLOUD_PROJECT}