Agent 定义方法——模具行业从业者必备知识手册

system 2

<h2>Agent 定义方法</h2>

<ul>
<li>LLM Agent: <code>import veagent "github.com/volcengine/veadk-go/agent/llmagent"</code></li>
<li>Sequential Agent: <code>import "github.com/volcengine/veadk-go/agent/workflowagents/sequentialagent"</code></li>
<li>Loop Agent: <code>import "github.com/volcengine/veadk-go/agent/workflowagents/loopagent"</code></li>
<li>Parallel Agent: <code>import "github.com/volcengine/veadk-go/agent/workflowagents/parallelagent"</code></li>
</ul>

<p>其中，LLM Agent 是最基础的智能体（由 LLM 启动进行自主决策），Sequential Agent 是按顺序执行的智能体，Loop Agent 是循环执行的智能体，Parallel Agent 是并行执行的智能体。</p>

<h4>1、你可以通过如下方式定义智能体：</h4>

<pre><code>
import (
"context"
"fmt"

veagent "github.com/volcengine/veadk-go/agent/llmagent"
"github.com/volcengine/veadk-go/apps"
"github.com/volcengine/veadk-go/apps/agentkit_server_app"
vetool "github.com/volcengine/veadk-go/tool"
"google.golang.org/adk/agent"
"google.golang.org/adk/agent/llmagent"
"google.golang.org/adk/tool"

)

func main() {
ctx := context.Background()

subAgent, err := veagent.New(&veagent.Config{
	Config: llmagent.Config{
		Name:        "...",
		Description: "...",
		Instruction: `...`,
	},
	ModelName: "...",
})
if err != nil {
	fmt.Printf("NewLLMAgent subAgent failed: %v", err)
	return
}

rootAgent, err := veagent.New(&veagent.Config{
	Config: llmagent.Config{
		Name:        "...",
		Description: "...",
		Instruction: `...`,
		SubAgents: []agent.Agent{subAgent},
	},
	ModelName: "...",
})
if err != nil {
	fmt.Printf("NewLLMAgent rootAgent failed: %v", err)
	return
}

app := agentkit_server_app.NewAgentkitServerApp(apps.DefaultApiConfig())

err = app.Run(ctx, &apps.RunConfig{
	AgentLoader: agent.NewSingleLoader(rootAgent),
})
if err != nil {
	fmt.Printf("Run failed: %v", err)
}

}

</code></pre>

<h4>2、可以生成一个强制按顺序执行的智能体：</h4>

<pre><code>
import (
"context"
"fmt"

veagent "github.com/volcengine/veadk-go/agent/llmagent"
"github.com/volcengine/veadk-go/agent/workflowagents/sequentialagent"
"github.com/volcengine/veadk-go/apps"
"github.com/volcengine/veadk-go/apps/agentkit_server_app"
"google.golang.org/adk/agent"
"google.golang.org/adk/agent/llmagent"

)

func main() {
ctx := context.Background()

agent1, err := veagent.New(&veagent.Config{
	Config: llmagent.Config{
		Name:        "...",
		Description: "...",
		Instruction: "...",
	},
})
if err != nil {
	fmt.Printf("NewLLMAgent agent1 failed: %v", err)
	return
}

agent2, err := veagent.New(&veagent.Config{
	Config: llmagent.Config{
		Name:        "...",
		Description: "...",
		Instruction: "...",
	},
})
if err != nil {
	fmt.Printf("NewLLMAgent agent failed: %v", err)
	return
}

rootAgent, err := sequentialagent.New(sequentialagent.Config{
	AgentConfig: agent.Config{
		Name:        "...",
		SubAgents:   []agent.Agent{agent1, agent2},
		Description: "...",
	},
})

if err != nil {
	fmt.Printf("NewSequentialAgent failed: %v", err)
	return
}

app := agentkit_server_app.NewAgentkitServerApp(apps.DefaultApiConfig())

err = app.Run(ctx, &apps.RunConfig{
	AgentLoader: agent.NewSingleLoader(rootAgent),
})
if err != nil {
	fmt.Printf("Run failed: %v", err)
}

}
</code></pre>

<p><code>agent1</code> 与 <code>agent2</code> 将会严格按顺序执行</p>

<p>注意，根智能体的命名必须为 <code>rootAgent</code>。</p>

<h3>让 Agent 结构化输出</h3>

<p>为保证更高的准确率和 Agent 执行时的可控性，使用结构化输出是一种有效的手段。</p>

<p>在定义 Agent 时，通过 <code>model_extra_config={"response_format": ...}</code> 可以让 Agent 结构化输出。其中，<code>...</code> 是你定义的 Pydantic 模型，用于描述 Agent 的输出格式。</p>

<pre><code>
from pydantic import BaseModel
from veadk import Agent, Runner

定义分步解析模型（对应业务场景的结构化响应）

class Step(BaseModel):
explanation: str # 步骤说明
output: str # 步骤计算结果

定义最终响应模型（包含分步过

各位师傅觉得这个方案怎么样？有更好的做法吗？
</code></pre>