reasoning

Reasoning Model Example / 推理模型示例

演示如何在 agno-Go 中使用推理模型。

Demonstrates how to use reasoning models in agno-Go.

支持的推理模型 / Supported Reasoning Models

OpenAI

• o1-preview
• o1-mini
• o3 系列
• o4 系列

Google Gemini

• gemini-2.5-flash-thinking
• 所有包含 "thinking" 关键词的模型

Anthropic Claude

• 需要显式配置 thinking 参数
• Requires explicit thinking configuration

运行示例 / Run Example

# 设置 API Key / Set API Key
export OPENAI_API_KEY=your-api-key

# 运行示例 / Run example
go run main.go

功能特性 / Features

• ✅ 自动检测: 自动识别推理模型

• ✅ 推理提取: 自动提取并存储推理内容

• ✅ 优雅降级: 提取失败不影响 Agent 执行

• ✅ 零配置: 无需额外配置,开箱即用

• ✅ 性能优化: 仅对推理模型执行提取操作

• ✅ Auto Detection: Automatically identifies reasoning models

• ✅ Reasoning Extraction: Automatically extracts and stores reasoning content

• ✅ Graceful Degradation: Extraction failures don't interrupt Agent execution

• ✅ Zero Configuration: Works out of the box

• ✅ Performance Optimized: Extraction only runs for reasoning models

示例输出 / Example Output

🤖 ReasoningAgent is thinking...
📝 Input: Solve this complex problem: ...

💬 Response:
[Model's final answer]

🧠 Reasoning Process:
[Model's step-by-step thinking process]

📊 Reasoning Tokens: 1250

代码说明 / Code Explanation

// 推理内容自动提取和存储
// Reasoning content is automatically extracted and stored
for _, msg := range output.Messages {
    if msg.ReasoningContent != nil {
        // 访问推理内容 / Access reasoning content
        fmt.Println(msg.ReasoningContent.Content)

        // 可选字段 / Optional fields
        if msg.ReasoningContent.TokenCount != nil {
            // Token 统计 / Token count
        }
        if msg.ReasoningContent.RedactedContent != nil {
            // 脱敏内容 / Redacted content
        }
    }
}

高级用法 / Advanced Usage

使用不同的推理模型 / Using Different Reasoning Models

// OpenAI o1-mini (更快更便宜)
model, _ := openaiModel.New(openaiModel.Config{
    Model: "o1-mini",
})

// Gemini 2.5 Flash Thinking
// (需要安装 Gemini 支持 / Requires Gemini support)
// model, _ := geminiModel.New(geminiModel.Config{
//     Model: "gemini-2.5-flash-thinking",
// })

访问完整推理历史 / Accessing Full Reasoning History

output, _ := agent.Run(ctx, input)

// 遍历所有消息,包括历史推理 / Iterate through all messages
for i, msg := range output.Messages {
    if msg.Role == types.RoleAssistant && msg.ReasoningContent != nil {
        fmt.Printf("Turn %d Reasoning:\n%s\n", i, msg.ReasoningContent.Content)
    }
}

注意事项 / Notes

1. API 费用: 推理模型通常比标准模型更昂贵

2. 响应时间: 推理模型可能需要更长的处理时间

3. Token 限制: 注意推理内容会占用额外的 token

4. API Costs: Reasoning models are typically more expensive

5. Response Time: Reasoning models may take longer to process

6. Token Limits: Be aware that reasoning content uses additional tokens