google-agents-cli

Skill by ara.so — Devtools Skills collection.

agents-cli

is the CLI and skills framework for building, evaluating, and deploying AI agents on Google Cloud's Gemini Enterprise Agent Platform. It works with ADK (Agent Development Kit) to provide end-to-end agent development workflows, from scaffolding to production deployment.

由ara.so提供的Skill——Devtools Skills合集。

agents-cli

是一款CLI工具及技能框架，用于在Google Cloud的Gemini Enterprise Agent平台上构建、评估和部署AI Agent。它与ADK（Agent Development Kit）配合使用，提供从项目搭建到生产部署的端到端Agent开发工作流。

What It Does

功能介绍

Scaffold agent projects: Create new ADK agent projects with best practices built-in
Local development: Run and test agents locally with hot reload
Evaluation: Run systematic evaluations with metrics, evalsets, and LLM-as-judge
Deployment: Deploy to Google Cloud (Agent Runtime, Cloud Run, GKE)
Publishing: Register agents with Gemini Enterprise
Observability: Integrate Cloud Trace, logging, and third-party monitoring
CI/CD: Set up staging/prod pipelines with automated testing

项目搭建: 创建内置最佳实践的全新ADK Agent项目
本地开发: 本地运行并测试Agent，支持热重载
评估: 通过指标、评估集和LLM-as-judge运行系统性评估
部署: 部署至Google Cloud（Agent Runtime、Cloud Run、GKE）
发布: 在Gemini Enterprise中注册Agent
可观测性: 集成Cloud Trace、日志记录及第三方监控
CI/CD: 搭建包含自动化测试的预发布/生产环境流水线

Installation

安装

Prerequisites: Python 3.11+, uv, Node.js

bash

undefined

前置要求: Python 3.11+、uv、Node.js

bash

undefined

Install CLI and skills

uvx google-agents-cli setup

Or just install the CLI

pip install google-agents-cli

Verify installation

agents-cli --version

undefined

agents-cli --version

undefined

Authentication

身份验证

bash

undefined

bash

undefined

Authenticate with Google Cloud

agents-cli login

Check authentication status

agents-cli login --status

For local development without Google Cloud, use AI Studio API key

export GOOGLE_API_KEY=your_api_key_here

undefined

export GOOGLE_API_KEY=your_api_key_here

undefined

Core Commands

核心命令

Project Scaffolding

项目搭建

bash

undefined

bash

undefined

Create a new agent project

agents-cli scaffold my-agent

Create with specific template

agents-cli scaffold my-agent --template basic agents-cli scaffold my-agent --template rag

Add features to existing project

agents-cli scaffold enhance --add deployment agents-cli scaffold enhance --add cicd agents-cli scaffold enhance --add rag

Upgrade project to newer agents-cli version

agents-cli scaffold upgrade

undefined

agents-cli scaffold upgrade

undefined

Local Development

本地开发

bash

undefined

bash

undefined

Install project dependencies

agents-cli install

Run agent with a single prompt

agents-cli run "What's the weather in San Francisco?"

Run with file input

agents-cli run --input-file prompt.txt

Run with streaming output

agents-cli run "Summarize this article" --stream

Lint code

agents-cli lint

undefined

agents-cli lint

undefined

Evaluation

评估

bash

undefined

bash

undefined

Run evaluations

agents-cli eval run

Run specific evalset

agents-cli eval run --evalset evalsets/basic.yaml

Compare two evaluation results

agents-cli eval compare results/eval1.json results/eval2.json

Run with custom metrics

agents-cli eval run --metrics accuracy,latency,cost

undefined

agents-cli eval run --metrics accuracy,latency,cost

undefined

Deployment

部署

bash

undefined

bash

undefined

Deploy to Google Cloud (interactive)

agents-cli deploy

Deploy with specific config

agents-cli deploy --config deploy.yaml

Deploy to specific environment

agents-cli deploy --env production

Provision infrastructure

agents-cli infra single-project --project-id my-project agents-cli infra cicd --project-id my-project

Set up datastore for RAG

agents-cli infra datastore --project-id my-project

undefined

agents-cli infra datastore --project-id my-project

undefined

Publishing

发布

bash

undefined

bash

undefined

Register with Gemini Enterprise

agents-cli publish gemini-enterprise

Publish with metadata

agents-cli publish gemini-enterprise --name "My Agent" --description "Does X"

undefined

agents-cli publish gemini-enterprise --name "My Agent" --description "Does X"

undefined

Data Ingestion (RAG)

数据摄入（RAG）

bash

undefined

bash

undefined

Run data ingestion pipeline

agents-cli data-ingestion --source gs://my-bucket/docs agents-cli data-ingestion --source ./local-docs --datastore my-datastore

undefined

agents-cli data-ingestion --source gs://my-bucket/docs agents-cli data-ingestion --source ./local-docs --datastore my-datastore

undefined

Utilities

实用工具

bash

undefined

bash

undefined

Show project info and CLI version

agents-cli info

Force reinstall skills to all IDEs

agents-cli update

undefined

agents-cli update

undefined

ADK Agent Code Patterns

ADK Agent代码模式

Basic Agent Structure

基础Agent结构

python

undefined

python

undefined

my_agent.py

from adk.agents import Agent from adk.tools import Tool from adk.models import ModelClient

Define a custom tool

class WeatherTool(Tool): """Get weather information for a location."""

def __init__(self):
    super().__init__(
        name="get_weather",
        description="Get current weather for a location"
    )

def execute(self, location: str) -> str:
    # Implementation
    return f"Weather in {location}: Sunny, 72°F"

class WeatherTool(Tool): """Get weather information for a location."""

def __init__(self):
    super().__init__(
        name="get_weather",
        description="Get current weather for a location"
    )

def execute(self, location: str) -> str:
    # Implementation
    return f"Weather in {location}: Sunny, 72°F"

Create agent

agent = Agent( name="weather-assistant", description="An agent that provides weather information", model=ModelClient(model_name="gemini-2.0-flash-exp"), tools=[WeatherTool()] )

Run agent

if name == "main": response = agent.run("What's the weather in NYC?") print(response)

undefined

if name == "main": response = agent.run("What's the weather in NYC?") print(response)

undefined

Agent with State Management

带状态管理的Agent

python

from adk.agents import Agent, AgentState
from adk.models import ModelClient
from typing import Any, Dict

class ConversationState(AgentState):
    """Custom state for conversation tracking."""
    
    def __init__(self):
        super().__init__()
        self.conversation_history = []
        self.user_preferences = {}
    
    def add_message(self, role: str, content: str):
        self.conversation_history.append({"role": role, "content": content})

agent = Agent(
    name="stateful-assistant",
    model=ModelClient(model_name="gemini-2.0-flash-exp"),
    state=ConversationState()
)

python

from adk.agents import Agent, AgentState
from adk.models import ModelClient
from typing import Any, Dict

class ConversationState(AgentState):
    """Custom state for conversation tracking."""
    
    def __init__(self):
        super().__init__()
        self.conversation_history = []
        self.user_preferences = {}
    
    def add_message(self, role: str, content: str):
        self.conversation_history.append({"role": role, "content": content})

agent = Agent(
    name="stateful-assistant",
    model=ModelClient(model_name="gemini-2.0-flash-exp"),
    state=ConversationState()
)

Use state in agent execution

response = agent.run("Remember my name is Alice") agent.state.user_preferences["name"] = "Alice"

undefined

response = agent.run("Remember my name is Alice") agent.state.user_preferences["name"] = "Alice"

undefined

Multi-Agent Orchestration

多Agent编排

python

from adk.agents import Agent, AgentOrchestrator
from adk.models import ModelClient

python

from adk.agents import Agent, AgentOrchestrator
from adk.models import ModelClient

Create specialized agents

research_agent = Agent( name="researcher", description="Researches topics and gathers information", model=ModelClient(model_name="gemini-2.0-flash-exp") )

writer_agent = Agent( name="writer", description="Writes content based on research", model=ModelClient(model_name="gemini-2.0-flash-exp") )

research_agent = Agent( name="researcher", description="Researches topics and gathers information", model=ModelClient(model_name="gemini-2.0-flash-exp") )

writer_agent = Agent( name="writer", description="Writes content based on research", model=ModelClient(model_name="gemini-2.0-flash-exp") )

Orchestrate agents

orchestrator = AgentOrchestrator( agents=[research_agent, writer_agent], workflow="sequential" # or "parallel", "conditional" )

Run orchestrated workflow

result = orchestrator.run("Write an article about AI agents")

undefined

result = orchestrator.run("Write an article about AI agents")

undefined

Agent with Callbacks

带回调的Agent

python

from adk.agents import Agent
from adk.callbacks import Callback
from adk.models import ModelClient

class LoggingCallback(Callback):
    """Log agent execution steps."""
    
    def on_agent_start(self, agent_name: str, input_data: Any):
        print(f"Agent {agent_name} starting with input: {input_data}")
    
    def on_tool_start(self, tool_name: str, tool_input: Dict[str, Any]):
        print(f"Tool {tool_name} called with: {tool_input}")
    
    def on_tool_end(self, tool_name: str, tool_output: Any):
        print(f"Tool {tool_name} returned: {tool_output}")
    
    def on_agent_end(self, agent_name: str, output: Any):
        print(f"Agent {agent_name} finished with: {output}")

agent = Agent(
    name="monitored-agent",
    model=ModelClient(model_name="gemini-2.0-flash-exp"),
    callbacks=[LoggingCallback()]
)

python

from adk.agents import Agent
from adk.callbacks import Callback
from adk.models import ModelClient

class LoggingCallback(Callback):
    """Log agent execution steps."""
    
    def on_agent_start(self, agent_name: str, input_data: Any):
        print(f"Agent {agent_name} starting with input: {input_data}")
    
    def on_tool_start(self, tool_name: str, tool_input: Dict[str, Any]):
        print(f"Tool {tool_name} called with: {tool_input}")
    
    def on_tool_end(self, tool_name: str, tool_output: Any):
        print(f"Tool {tool_name} returned: {tool_output}")
    
    def on_agent_end(self, agent_name: str, output: Any):
        print(f"Agent {agent_name} finished with: {output}")

agent = Agent(
    name="monitored-agent",
    model=ModelClient(model_name="gemini-2.0-flash-exp"),
    callbacks=[LoggingCallback()]
)

RAG Agent Pattern

RAG Agent模式

python

from adk.agents import Agent
from adk.tools import Tool
from adk.models import ModelClient
from adk.rag import VectorStore, Retriever

class RAGTool(Tool):
    """Retrieve relevant documents from vector store."""
    
    def __init__(self, datastore_id: str):
        super().__init__(
            name="retrieve_docs",
            description="Retrieve relevant documents"
        )
        self.retriever = Retriever(datastore_id=datastore_id)
    
    def execute(self, query: str) -> str:
        docs = self.retriever.retrieve(query, top_k=5)
        return "\n\n".join([doc.content for doc in docs])

agent = Agent(
    name="rag-assistant",
    description="Agent with RAG capabilities",
    model=ModelClient(model_name="gemini-2.0-flash-exp"),
    tools=[RAGTool(datastore_id="my-datastore")]
)

python

from adk.agents import Agent
from adk.tools import Tool
from adk.models import ModelClient
from adk.rag import VectorStore, Retriever

class RAGTool(Tool):
    """Retrieve relevant documents from vector store."""
    
    def __init__(self, datastore_id: str):
        super().__init__(
            name="retrieve_docs",
            description="Retrieve relevant documents"
        )
        self.retriever = Retriever(datastore_id=datastore_id)
    
    def execute(self, query: str) -> str:
        docs = self.retriever.retrieve(query, top_k=5)
        return "\n\n".join([doc.content for doc in docs])

agent = Agent(
    name="rag-assistant",
    description="Agent with RAG capabilities",
    model=ModelClient(model_name="gemini-2.0-flash-exp"),
    tools=[RAGTool(datastore_id="my-datastore")]
)

Project Configuration

项目配置

agents.yaml

agents.yaml

yaml

undefined

yaml

undefined

Project configuration

Agent configuration

agent: name: my-assistant model: gemini-2.0-flash-exp temperature: 0.7 max_tokens: 2048

Tools configuration

tools:

name: web_search enabled: true
name: code_execution enabled: false

tools:

name: web_search enabled: true
name: code_execution enabled: false

Evaluation configuration

evaluation: evalsets: - path: evalsets/basic.yaml - path: evalsets/advanced.yaml metrics: - accuracy - latency - cost

Deployment configuration

deployment: target: cloud-run region: us-central1 min_instances: 1 max_instances: 10

Observability

observability: cloud_trace: true cloud_logging: true

undefined

observability: cloud_trace: true cloud_logging: true

undefined

Evalset Configuration

评估集配置

yaml

undefined

yaml

undefined

evalsets/basic.yaml

test_cases:

id: tc-001 input: "What is 2+2?" expected_output: "4" metrics:
- accuracy
- latency
id: tc-002 input: "Explain quantum computing in simple terms" judge: type: llm-as-judge criteria: - clarity - accuracy - conciseness
id: tc-003 input: "Write a Python function to reverse a string" validator: type: code-execution test: | def test_reverse(): assert reverse("hello") == "olleh" assert reverse("") == ""

undefined

test_cases:

id: tc-001 input: "What is 2+2?" expected_output: "4" metrics:
- accuracy
- latency
id: tc-002 input: "Explain quantum computing in simple terms" judge: type: llm-as-judge criteria: - clarity - accuracy - conciseness
id: tc-003 input: "Write a Python function to reverse a string" validator: type: code-execution test: | def test_reverse(): assert reverse("hello") == "olleh" assert reverse("") == ""

undefined

Deployment Configuration

部署配置

yaml

undefined

yaml

undefined

deploy.yaml

target: cloud-run project_id: ${GCP_PROJECT_ID} region: us-central1

service: name: my-agent-service min_instances: 1 max_instances: 10 cpu: 2 memory: 4Gi timeout: 300s

environment:

name: GOOGLE_API_KEY secret: projects/${GCP_PROJECT_ID}/secrets/gemini-api-key
name: LOG_LEVEL value: INFO

ci_cd: enabled: true environments: - name: staging project_id: ${STAGING_PROJECT_ID} branch: develop - name: production project_id: ${PRODUCTION_PROJECT_ID} branch: main

undefined

target: cloud-run project_id: ${GCP_PROJECT_ID} region: us-central1

service: name: my-agent-service min_instances: 1 max_instances: 10 cpu: 2 memory: 4Gi timeout: 300s

environment:

name: GOOGLE_API_KEY secret: projects/${GCP_PROJECT_ID}/secrets/gemini-api-key
name: LOG_LEVEL value: INFO

ci_cd: enabled: true environments: - name: staging project_id: ${STAGING_PROJECT_ID} branch: develop - name: production project_id: ${PRODUCTION_PROJECT_ID} branch: main

undefined

Environment Variables

环境变量

bash

undefined

bash

undefined

API Keys

export GOOGLE_API_KEY=your_api_key_here export GOOGLE_CLOUD_PROJECT=your-project-id

Agent Configuration

export AGENT_MODEL=gemini-2.0-flash-exp export AGENT_TEMPERATURE=0.7 export AGENT_MAX_TOKENS=2048

Deployment

export DEPLOY_REGION=us-central1 export DEPLOY_ENV=production

Observability

export ENABLE_CLOUD_TRACE=true export ENABLE_CLOUD_LOGGING=true export LOG_LEVEL=INFO

RAG

export DATASTORE_ID=my-datastore export VECTOR_STORE_TYPE=vertex-ai-search

undefined

export DATASTORE_ID=my-datastore export VECTOR_STORE_TYPE=vertex-ai-search

undefined

Common Patterns

常见模式

Creating a Complete Agent Project

创建完整的Agent项目

bash

undefined

bash

undefined

1. Scaffold project

agents-cli scaffold my-agent

2. Navigate to project

cd my-agent

3. Install dependencies

agents-cli install

4. Edit agent code

Edit src/agent.py with your agent logic

5. Run locally

agents-cli run "Test prompt"

6. Create evalset

Create evalsets/test.yaml

7. Run evaluations

agents-cli eval run

8. Deploy to Google Cloud

agents-cli login agents-cli deploy

9. Publish to Gemini Enterprise

agents-cli publish gemini-enterprise

undefined

agents-cli publish gemini-enterprise

undefined

Adding Tools to an Agent

为Agent添加工具

python

from adk.agents import Agent
from adk.tools import Tool
from adk.models import ModelClient
import requests

class SearchTool(Tool):
    """Search the web for information."""
    
    def __init__(self, api_key: str):
        super().__init__(
            name="web_search",
            description="Search the web for current information"
        )
        self.api_key = api_key
    
    def execute(self, query: str, num_results: int = 5) -> str:
        # Implementation using search API
        # Use self.api_key from environment
        results = self._search(query, num_results)
        return "\n".join([r["title"] + ": " + r["snippet"] for r in results])
    
    def _search(self, query: str, num_results: int):
        # Actual search implementation
        pass

class CalculatorTool(Tool):
    """Perform mathematical calculations."""
    
    def __init__(self):
        super().__init__(
            name="calculator",
            description="Perform mathematical calculations"
        )
    
    def execute(self, expression: str) -> float:
        # Safe evaluation of mathematical expressions
        import ast
        import operator
        
        operators = {
            ast.Add: operator.add,
            ast.Sub: operator.sub,
            ast.Mult: operator.mul,
            ast.Div: operator.truediv,
        }
        
        def eval_expr(node):
            if isinstance(node, ast.Num):
                return node.n
            elif isinstance(node, ast.BinOp):
                return operators[type(node.op)](
                    eval_expr(node.left),
                    eval_expr(node.right)
                )
            else:
                raise ValueError("Unsupported expression")
        
        return eval_expr(ast.parse(expression, mode='eval').body)

python

from adk.agents import Agent
from adk.tools import Tool
from adk.models import ModelClient
import requests

class SearchTool(Tool):
    """Search the web for information."""
    
    def __init__(self, api_key: str):
        super().__init__(
            name="web_search",
            description="Search the web for current information"
        )
        self.api_key = api_key
    
    def execute(self, query: str, num_results: int = 5) -> str:
        # Implementation using search API
        # Use self.api_key from environment
        results = self._search(query, num_results)
        return "\n".join([r["title"] + ": " + r["snippet"] for r in results])
    
    def _search(self, query: str, num_results: int):
        # Actual search implementation
        pass

class CalculatorTool(Tool):
    """Perform mathematical calculations."""
    
    def __init__(self):
        super().__init__(
            name="calculator",
            description="Perform mathematical calculations"
        )
    
    def execute(self, expression: str) -> float:
        # Safe evaluation of mathematical expressions
        import ast
        import operator
        
        operators = {
            ast.Add: operator.add,
            ast.Sub: operator.sub,
            ast.Mult: operator.mul,
            ast.Div: operator.truediv,
        }
        
        def eval_expr(node):
            if isinstance(node, ast.Num):
                return node.n
            elif isinstance(node, ast.BinOp):
                return operators[type(node.op)](
                    eval_expr(node.left),
                    eval_expr(node.right)
                )
            else:
                raise ValueError("Unsupported expression")
        
        return eval_expr(ast.parse(expression, mode='eval').body)

Create agent with multiple tools

import os

agent = Agent( name="research-assistant", model=ModelClient(model_name="gemini-2.0-flash-exp"), tools=[ SearchTool(api_key=os.getenv("SEARCH_API_KEY")), CalculatorTool() ] )

undefined

import os

agent = Agent( name="research-assistant", model=ModelClient(model_name="gemini-2.0-flash-exp"), tools=[ SearchTool(api_key=os.getenv("SEARCH_API_KEY")), CalculatorTool() ] )

undefined

Setting Up CI/CD

搭建CI/CD

bash

undefined

bash

undefined

1. Set up CI/CD infrastructure

agents-cli infra cicd --project-id my-project

2. Create deployment configs for staging and prod

Edit deploy-staging.yaml and deploy-production.yaml

3. Configure GitHub Actions or Cloud Build

The scaffold enhance command creates templates

4. Push to repository

git add . git commit -m "Set up CI/CD" git push origin main

undefined

git add . git commit -m "Set up CI/CD" git push origin main

undefined

Implementing Custom Metrics

实现自定义指标

python

undefined

python

undefined

evalsets/custom_metrics.py

from adk.evaluation import Metric from typing import Dict, Any

class CustomAccuracyMetric(Metric): """Custom accuracy metric with fuzzy matching."""

def __init__(self, threshold: float = 0.8):
    super().__init__(name="custom_accuracy")
    self.threshold = threshold

def evaluate(self, 
             prediction: str, 
             expected: str, 
             context: Dict[str, Any]) -> float:
    # Custom evaluation logic
    from difflib import SequenceMatcher
    ratio = SequenceMatcher(None, prediction, expected).ratio()
    return 1.0 if ratio >= self.threshold else 0.0

class LatencyMetric(Metric): """Measure agent response latency."""

def __init__(self):
    super().__init__(name="latency")

def evaluate(self,
             prediction: str,
             expected: str,
             context: Dict[str, Any]) -> float:
    # Return latency in milliseconds
    return context.get("execution_time_ms", 0)

from adk.evaluation import Metric from typing import Dict, Any

class CustomAccuracyMetric(Metric): """Custom accuracy metric with fuzzy matching."""

def __init__(self, threshold: float = 0.8):
    super().__init__(name="custom_accuracy")
    self.threshold = threshold

def evaluate(self, 
             prediction: str, 
             expected: str, 
             context: Dict[str, Any]) -> float:
    # Custom evaluation logic
    from difflib import SequenceMatcher
    ratio = SequenceMatcher(None, prediction, expected).ratio()
    return 1.0 if ratio >= self.threshold else 0.0

class LatencyMetric(Metric): """Measure agent response latency."""

def __init__(self):
    super().__init__(name="latency")

def evaluate(self,
             prediction: str,
             expected: str,
             context: Dict[str, Any]) -> float:
    # Return latency in milliseconds
    return context.get("execution_time_ms", 0)

Use in evaluation

from adk.evaluation import Evaluator

evaluator = Evaluator( agent=my_agent, evalset_path="evalsets/basic.yaml", metrics=[ CustomAccuracyMetric(threshold=0.85), LatencyMetric() ] )

results = evaluator.run() print(f"Average accuracy: {results.metrics['custom_accuracy']}") print(f"Average latency: {results.metrics['latency']}ms")

undefined

from adk.evaluation import Evaluator

evaluator = Evaluator( agent=my_agent, evalset_path="evalsets/basic.yaml", metrics=[ CustomAccuracyMetric(threshold=0.85), LatencyMetric() ] )

results = evaluator.run() print(f"Average accuracy: {results.metrics['custom_accuracy']}") print(f"Average latency: {results.metrics['latency']}ms")

undefined

Troubleshooting

故障排查

Authentication Issues

身份验证问题

bash

undefined

bash

undefined

Clear credentials and re-authenticate

agents-cli login --force

Use service account for CI/CD

export GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json agents-cli login --service-account

Check authentication status

agents-cli login --status

undefined

agents-cli login --status

undefined

Deployment Failures

部署失败

bash

undefined

bash

undefined

Check project configuration

agents-cli info

Validate deployment config

agents-cli deploy --validate-only

Deploy with verbose logging

agents-cli deploy --verbose

Check Cloud Run logs

gcloud run services logs read my-agent-service --project=${GCP_PROJECT_ID}

undefined

gcloud run services logs read my-agent-service --project=${GCP_PROJECT_ID}

undefined

Evaluation Issues

评估问题

bash

undefined

bash

undefined

Run single test case

agents-cli eval run --test-case tc-001

Run with debug mode

agents-cli eval run --debug

Validate evalset format

agents-cli eval run --validate-only

Check evaluation logs

cat .agents-cli/eval-results/latest.log

undefined

cat .agents-cli/eval-results/latest.log

undefined

Dependency Conflicts

依赖冲突

bash

undefined

bash

undefined

Reinstall dependencies

agents-cli install --force

Clear cache

rm -rf .agents-cli/cache agents-cli install

Use specific Python version

uv venv --python 3.11 source .venv/bin/activate agents-cli install

undefined

uv venv --python 3.11 source .venv/bin/activate agents-cli install

undefined

Model Access Issues

模型访问问题

bash

undefined

bash

undefined

Verify API key is set

echo $GOOGLE_API_KEY

Test model access directly

curl -X POST "https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash-exp:generateContent?key=${GOOGLE_API_KEY}"
-H 'Content-Type: application/json'
-d '{"contents":[{"parts":[{"text":"Hello"}]}]}'

Switch to different model

export AGENT_MODEL=gemini-1.5-pro agents-cli run "test"

undefined

export AGENT_MODEL=gemini-1.5-pro agents-cli run "test"

undefined

Performance Optimization

性能优化

python

undefined

python

undefined

Enable caching for repeated queries

from adk.models import ModelClient

model = ModelClient( model_name="gemini-2.0-flash-exp", cache_enabled=True, cache_ttl=3600 # 1 hour )

from adk.models import ModelClient

model = ModelClient( model_name="gemini-2.0-flash-exp", cache_enabled=True, cache_ttl=3600 # 1 hour )

Batch processing for evaluations

agents-cli eval run --batch-size 10 --parallel 4

Reduce token usage

agent = Agent( name="efficient-agent", model=ModelClient( model_name="gemini-2.0-flash-exp", max_tokens=1024, # Lower limit temperature=0.3 # More deterministic ) )

undefined

agent = Agent( name="efficient-agent", model=ModelClient( model_name="gemini-2.0-flash-exp", max_tokens=1024, # Lower limit temperature=0.3 # More deterministic ) )

undefined

Debugging Agent Behavior

调试Agent行为

python

from adk.agents import Agent
from adk.callbacks import DebugCallback
from adk.models import ModelClient

python

from adk.agents import Agent
from adk.callbacks import DebugCallback
from adk.models import ModelClient

Enable detailed debugging

debug_callback = DebugCallback( log_prompts=True, log_responses=True, log_tool_calls=True )

agent = Agent( name="debug-agent", model=ModelClient(model_name="gemini-2.0-flash-exp"), callbacks=[debug_callback], verbose=True )

debug_callback = DebugCallback( log_prompts=True, log_responses=True, log_tool_calls=True )

agent = Agent( name="debug-agent", model=ModelClient(model_name="gemini-2.0-flash-exp"), callbacks=[debug_callback], verbose=True )

Check agent execution traces

response = agent.run("Debug this behavior") print(agent.get_trace())

undefined

response = agent.run("Debug this behavior") print(agent.get_trace())

undefined

Model Selection Guide

模型选择指南

python

undefined

python

undefined

Fast, cost-effective for simple tasks

ModelClient(model_name="gemini-2.0-flash-exp")

Balanced performance and capabilities

ModelClient(model_name="gemini-1.5-pro")

Maximum capabilities for complex reasoning

ModelClient(model_name="gemini-1.5-pro-002")

Multimodal tasks (images, video)

ModelClient( model_name="gemini-1.5-pro-vision", multimodal=True )

undefined

ModelClient( model_name="gemini-1.5-pro-vision", multimodal=True )

undefined

google-agents-cli

Original

Translation

google-agents-cli

google-agents-cli

What It Does

功能介绍

Installation

安装

Install CLI and skills

Install CLI and skills

Or just install the CLI

Or just install the CLI

Verify installation

Verify installation

Authentication

身份验证

Authenticate with Google Cloud

Authenticate with Google Cloud

Check authentication status

Check authentication status

For local development without Google Cloud, use AI Studio API key

For local development without Google Cloud, use AI Studio API key

Core Commands

核心命令

Project Scaffolding

项目搭建

Create a new agent project

Create a new agent project

Create with specific template

Create with specific template

Add features to existing project

Add features to existing project

Upgrade project to newer agents-cli version

Upgrade project to newer agents-cli version

Local Development

本地开发

Install project dependencies

Install project dependencies

Run agent with a single prompt

Run agent with a single prompt

Run with file input

Run with file input

Run with streaming output

Run with streaming output

Lint code

Lint code

Evaluation

评估

Run evaluations

Run evaluations

Run specific evalset

Run specific evalset

Compare two evaluation results

Compare two evaluation results

Run with custom metrics

Run with custom metrics

Deployment

部署

Deploy to Google Cloud (interactive)

Deploy to Google Cloud (interactive)

Deploy with specific config

Deploy with specific config

Deploy to specific environment

Deploy to specific environment

Provision infrastructure

Provision infrastructure

Set up datastore for RAG

Set up datastore for RAG

Publishing

发布

Register with Gemini Enterprise

Register with Gemini Enterprise

Publish with metadata

Publish with metadata

Data Ingestion (RAG)

数据摄入（RAG）

Run data ingestion pipeline

Run data ingestion pipeline

Utilities

`agents.yaml`

`agents.yaml`