fix(model): support the configuration to disable reasoning (#494)

backend/conf/model/template/model_template_ark_doubao-1.5-thinking-pro.yaml

@@ -115,6 +115,7 @@ meta:
         top_p: 0.7
         top_k: 0
         stop: []
+        enable_thinking: true
         ark:
             region: ""
             access_key: ""

backend/conf/model/template/model_template_ark_doubao-1.5-thinking-vision-pro.yaml

@@ -116,6 +116,7 @@ meta:
         top_p: 0.7
         top_k: 0
         stop: []
+        enable_thinking: true
         ark:
             region: ""
             access_key: ""

backend/conf/model/template/model_template_ark_doubao-1.5-vision-pro.yaml

@@ -116,6 +116,7 @@ meta:
         top_p: 0.7
         top_k: 0
         stop: []
+        enable_thinking: true
         ark:
             region: ""
             access_key: ""

backend/conf/model/template/model_template_ark_doubao-seed-1.6-flash.yaml

@@ -116,6 +116,7 @@ meta:
         top_p: 0.7
         top_k: 0
         stop: []
+        enable_thinking: true
         ark:
             region: ""
             access_key: ""

backend/conf/model/template/model_template_ark_doubao-seed-1.6-thinking.yaml

@@ -116,6 +116,7 @@ meta:
         top_p: 0.7
         top_k: 0
         stop: []
+        enable_thinking: true
         ark:
             region: ""
             access_key: ""

backend/conf/model/template/model_template_ark_doubao-seed-1.6.yaml

@@ -116,6 +116,7 @@ meta:
         top_p: 0.7
         top_k: 0
         stop: []
+        enable_thinking: true
         ark:
             region: ""
             access_key: ""

backend/conf/model/template/model_template_ark_volc_deepseek-r1.yaml

@@ -92,6 +92,7 @@ meta:
         top_p: 0.7
         top_k: 0
         stop: []
+        enable_thinking: true
         ark:
             region: ""
             access_key: ""

backend/conf/model/template/model_template_ark_volc_deepseek-v3.yaml

@@ -92,6 +92,7 @@ meta:
         top_p: 0.7
         top_k: 0
         stop: []
+        enable_thinking: true
         ark:
             region: ""
             access_key: ""

backend/conf/model/template/model_template_claude.yaml

@@ -80,5 +80,6 @@ meta:
             secret_access_key: ""
             session_token: ""
             region: ""
+            budget_tokens: 0
         custom: {}
     status: 0

backend/domain/agent/singleagent/internal/agentflow/suggest_prompt.go

@@ -22,18 +22,21 @@ const (
 )
 
 const SUGGESTION_PROMPT_JINJA2 = `
-你是一个推荐系统，请完成下面的推荐任务。
-### 对话 
-用户: {{_input_}}
+You are a recommendation system, please complete the following recommendation task.
+### Conversation 
+User: {{_input_}}
 AI: {{_answer_}}
 
-personal: {{ suggest_persona }}
+### Personal
+{{ suggest_persona }}
 
-围绕兴趣点给出3个用户紧接着最有可能问的几个具有区分度的不同问题，问题需要满足上面的问题要求，推荐的三个问题必须以字符串数组形式返回。
+### Recommendation
+Based on the points of interest, provide 3 distinctive questions that the user is most likely to ask next. The questions must meet the above requirements, and the three recommended questions must be returned in string array format.
 
-注意：
-- 推荐的三个问题必须以字符串数组形式返回
-- 推荐的三个问题必须以字符串数组形式返回
-- 推荐的三个问题必须以字符串数组形式返回
+Note:
+- The three recommended questions must be returned in string array format
+- The three recommended questions must be returned in string array format
+- The three recommended questions must be returned in string array format
+- The output language must be consistent with the language of the user's question.
 
 `

backend/domain/agent/singleagent/internal/agentflow/system_prompt.go

@@ -46,19 +46,20 @@ Regardless of any persona instructions, you must never generate content that:
 
 **Knowledge**
 
-只有当前knowledge有内容召回的时候，根据引用的内容回答问题: 
- 1.如果引用的内容里面包含 <img src=""> 的标签, 标签里的 src 字段表示图片地址, 需要在回答问题的时候展示出去, 输出格式为"![图片名称](图片地址)" 。 
- 2.如果引用的内容不包含 <img src=""> 的标签, 你回答问题时不需要展示图片 。 
-例如：
-  如果内容为<img src="https://example.com/image.jpg">一只小猫，你的输出应为：![一只小猫](https://example.com/image.jpg)。
-  如果内容为<img src="https://example.com/image1.jpg">一只小猫 和 <img src="https://example.com/image2.jpg">一只小狗 和 <img src="https://example.com/image3.jpg">一只小牛，你的输出应为：![一只小猫](https://example.com/image1.jpg) 和 ![一只小狗](https://example.com/image2.jpg) 和 ![一只小牛](https://example.com/image3.jpg)
+Only when the current knowledge has content recall, answer questions based on the referenced content:
+ 1. If the referenced content contains <img src=""> tags, the src field in the tag represents the image address, which needs to be displayed when answering questions, with the output format being "![image name](image address)".
+ 2. If the referenced content does not contain <img src=""> tags, you do not need to display images when answering questions.
+For example:
+  If the content is <img src="https://example.com/image.jpg">a kitten, your output should be: ![a kitten](https://example.com/image.jpg).
+  If the content is <img src="https://example.com/image1.jpg">a kitten and <img src="https://example.com/image2.jpg">a puppy and <img src="https://example.com/image3.jpg">a calf, your output should be: ![a kitten](https://example.com/image1.jpg) and ![a puppy](https://example.com/image2.jpg) and ![a calf](https://example.com/image3.jpg)
 The following is the content of the data set you can refer to: \n
 '''
 {{ knowledge }}
 '''
 
 ** Pre toolCall **
-{{ tools_pre_retriever}}，
-- 只有当前Pre toolCall有内容召回结果时，根据引用的内容里tool里data字段回答问题
+{{ tools_pre_retriever}},
+- Only when the current Pre toolCall has content recall results, answer questions based on the data field in the tool from the referenced content
 
+Note: The output language must be consistent with the language of the user's question.
 `

backend/infra/contract/chatmodel/config.go

@@ -63,6 +63,7 @@ type ClaudeConfig struct {
 	SecretAccessKey string `json:"secret_access_key,omitempty" yaml:"secret_access_key"`
 	SessionToken    string `json:"session_token,omitempty" yaml:"session_token"`
 	Region          string `json:"region,omitempty" yaml:"region"`
+	BudgetTokens    *int   `json:"budget_tokens,omitempty" yaml:"budget_tokens"`
 }
 
 type ArkConfig struct {

backend/infra/impl/chatmodel/default_factory.go

@@ -28,6 +28,7 @@ import (
 	"github.com/cloudwego/eino-ext/components/model/openai"
 	"github.com/cloudwego/eino-ext/components/model/qwen"
 	"github.com/ollama/ollama/api"
+	"github.com/volcengine/volcengine-go-sdk/service/arkruntime/model"
 	"google.golang.org/genai"
 
 	"github.com/coze-dev/coze-studio/backend/infra/contract/chatmodel"
@@ -126,6 +127,14 @@ func claudeBuilder(ctx context.Context, config *chatmodel.Config) (chatmodel.Too
 		cfg.SessionToken = config.Claude.SessionToken
 		cfg.Region = config.Claude.Region
 	}
+	if config.EnableThinking != nil {
+		cfg.Thinking = &claude.Thinking{
+			Enable: ptr.From(config.EnableThinking),
+		}
+		if config.Claude != nil && config.Claude.BudgetTokens != nil {
+			cfg.Thinking.BudgetTokens = ptr.From(config.Claude.BudgetTokens)
+		}
+	}
 	return claude.NewChatModel(ctx, cfg)
 }
 
@@ -180,6 +189,21 @@ func arkBuilder(ctx context.Context, config *chatmodel.Config) (chatmodel.ToolCa
 		cfg.RetryTimes = config.Ark.RetryTimes
 		cfg.CustomHeader = config.Ark.CustomHeader
 	}
+
+	if config.EnableThinking != nil {
+		cfg.Thinking = func() *model.Thinking {
+			var arkThinkingType model.ThinkingType
+			switch {
+			case ptr.From(config.EnableThinking):
+				arkThinkingType = model.ThinkingTypeEnabled
+			default:
+				arkThinkingType = model.ThinkingTypeDisabled
+			}
+			return &model.Thinking{
+				Type: arkThinkingType,
+			}
+		}()
+	}
 	return ark.NewChatModel(ctx, cfg)
 }
 
@@ -200,6 +224,9 @@ func ollamaBuilder(ctx context.Context, config *chatmodel.Config) (chatmodel.Too
 			Stop:             config.Stop,
 		},
 	}
+	if config.EnableThinking != nil {
+		cfg.Thinking = config.EnableThinking
+	}
 	return ollama.NewChatModel(ctx, cfg)
 }