AI Agent plugin adds JSON formatting output feature (#1374)

2026-02-26 21:50:53 +08:00 · 2024-11-11 11:11:02 +08:00
parent 7697af9d2b
commit 39b6eac9d0
5 changed files with 355 additions and 67 deletions
--- a/plugins/wasm-go/extensions/ai-agent/README.md
+++ b/plugins/wasm-go/extensions/ai-agent/README.md
@@ -5,7 +5,7 @@ description: AI Agent插件配置参考
 ---

 ## 功能说明
-一个可定制化的 API AI Agent，支持配置 http method 类型为 GET 与 POST 的 API，支持多轮对话，支持流式与非流式模式。
+一个可定制化的 API AI Agent，支持配置 http method 类型为 GET 与 POST 的 API，支持多轮对话，支持流式与非流式模式，支持将结果格式化为自定义的 json。
 agent流程图如下：
 ![ai-agent](https://img.alicdn.com/imgextra/i1/O1CN01PGSDW31WQfEPm173u_!!6000000002783-0-tps-2733-1473.jpg)

@@ -21,6 +21,7 @@ agent流程图如下：
 | `llm`            | object    | 必填    | -      | 配置 AI 服务提供商的信息     |
 | `apis`           | object    | 必填    | -      | 配置外部 API 服务提供商的信息 |
 | `promptTemplate` | object    | 非必填  | -      | 配置 Agent ReAct 模板的信息  |
+| `jsonResp`       | object    | 非必填  | -      | 配置 json 格式化的相关信息   |

 `llm`的配置字段说明如下：

@@ -78,7 +79,14 @@ agent流程图如下：
 | `observation`   | string    | 非必填     | -      | Agent ReAct 模板的 observation 部分          |
 | `thought2`      | string    | 非必填     | -      | Agent ReAct 模板的 thought2 部分             |

-## 用法示例
+`jsonResp`的配置字段说明如下：
+
+| 名称               | 数据类型   | 填写要求 | 默认值 | 描述                               |
+|--------------------|-----------|---------|--------|-----------------------------------|
+| `enable`           | bool      | 非必填   | false  | 是否开启 json 格式化。             |
+| `jsonSchema`       | string    | 非必填   | -      | 自定义 json schema                |
+
+## 用法示例-不开启 json 格式化

 **配置信息**

@@ -293,7 +301,7 @@ deepl提供了一个工具，用于翻译给定的句子，支持多语言。。
 **请求示例**

 ```shell
-curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
+curl 'http://<这里换成网关地址>/api/openai/v1/chat/completions' \
 -H 'Accept: application/json, text/event-stream' \
 -H 'Content-Type: application/json' \
 --data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role":"user","content":"我想在济南市鑫盛大厦附近喝咖啡，给我推荐几个"}],"presence_penalty":0,"temperature":0,"top_p":0}'
@@ -308,7 +316,7 @@ curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
 **请求示例**

 ```shell
-curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
+curl 'http://<这里换成网关地址>/api/openai/v1/chat/completions' \
 -H 'Accept: application/json, text/event-stream' \
 -H 'Content-Type: application/json' \
 --data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role":"user","content":"济南市现在的天气情况如何？"}],"presence_penalty":0,"temperature":0,"top_p":0}'
@@ -323,7 +331,7 @@ curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
 **请求示例**

 ```shell
-curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
+curl 'http://<这里换成网关地址>/api/openai/v1/chat/completions' \
 -H 'Accept: application/json, text/event-stream' \
 -H 'Content-Type: application/json' \
 --data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role": "user","content": "济南的天气如何？"},{ "role": "assistant","content": "目前，济南市的天气为多云，气温为24℃，数据更新时间为2024年9月12日21时50分14秒。"},{"role": "user","content": "北京呢？"}],"presence_penalty":0,"temperature":0,"top_p":0}'
@@ -338,7 +346,7 @@ curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
 **请求示例**

 ```shell
-curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
+curl 'http://<这里换成网关地址>/api/openai/v1/chat/completions' \
 -H 'Accept: application/json, text/event-stream' \
 -H 'Content-Type: application/json' \
 --data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role":"user","content":"济南市现在的天气情况如何？用华氏度表示，用日语回答"}],"presence_penalty":0,"temperature":0,"top_p":0}'
@@ -353,7 +361,7 @@ curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
 **请求示例**

 ```shell
-curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
+curl 'http://<这里换成网关地址>/api/openai/v1/chat/completions' \
 -H 'Accept: application/json, text/event-stream' \
 -H 'Content-Type: application/json' \
 --data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role":"user","content":"帮我用德语翻译以下句子：九头蛇万岁!"}],"presence_penalty":0,"temperature":0,"top_p":0}'
@@ -364,3 +372,71 @@ curl 'http://<这里换成网关公网IP>/api/openai/v1/chat/completions' \
 ```json
 {"id":"65dcf12c-61ff-9e68-bffa-44fc9e6070d5","choices":[{"index":0,"message":{"role":"assistant","content":" “九头蛇万岁!”的德语翻译为“Hoch lebe Hydra!”。"},"finish_reason":"stop"}],"created":1724043865,"model":"qwen-max-0403","object":"chat.completion","usage":{"prompt_tokens":908,"completion_tokens":52,"total_tokens":960}}
 ```
+
+## 用法示例-开启 json 格式化
+
+**配置信息**
+在上述配置的基础上增加 jsonResp 配置
+```yaml
+jsonResp:
+  enable: true
+```
+
+**请求示例**
+
+```shell
+curl 'http://<这里换成网关地址>/api/openai/v1/chat/completions' \
+-H 'Accept: application/json, text/event-stream' \
+-H 'Content-Type: application/json' \
+--data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role":"user","content":"北京市现在的天气情况如何？"}],"presence_penalty":0,"temperature":0,"top_p":0}'
+```
+
+**响应示例**
+
+```json
+{"id":"ebd6ea91-8e38-9e14-9a5b-90178d2edea4","choices":[{"index":0,"message":{"role":"assistant","content": "{\"city\": \"北京市\", \"weather_condition\": \"多云\", \"temperature\": \"19℃\", \"data_update_time\": \"2024年10月9日16时37分53秒\"}"},"finish_reason":"stop"}],"created":1723187991,"model":"qwen-max-0403","object":"chat.completion","usage":{"prompt_tokens":890,"completion_tokens":56,"total_tokens":946}}
+```
+如果不自定义 json schema，大模型会自动生成一个 json 格式
+
+**配置信息**
+增加自定义 json schema 配置
+```yaml
+jsonResp:
+  enable: true
+  jsonSchema: |
+    title: WeatherSchema
+    type: object
+    properties:
+      location:
+        type: string
+        description: 城市名称.
+      weather:
+        type: string
+        description: 天气情况.
+      temperature:
+        type: string
+        description: 温度.
+      update_time:
+        type: string
+        description: 数据更新时间.
+    required:
+      - location
+      - weather
+      - temperature
+    additionalProperties: false
+```
+
+**请求示例**
+
+```shell
+curl 'http://<这里换成网关地址>/api/openai/v1/chat/completions' \
+-H 'Accept: application/json, text/event-stream' \
+-H 'Content-Type: application/json' \
+--data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role":"user","content":"北京市现在的天气情况如何？"}],"presence_penalty":0,"temperature":0,"top_p":0}'
+```
+
+**响应示例**
+
+```json
+{"id":"ebd6ea91-8e38-9e14-9a5b-90178d2edea4","choices":[{"index":0,"message":{"role":"assistant","content": "{\"location\": \"北京市\", \"weather\": \"多云\", \"temperature\": \"19℃\", \"update_time\": \"2024年10月9日16时37分53秒\"}"},"finish_reason":"stop"}],"created":1723187991,"model":"qwen-max-0403","object":"chat.completion","usage":{"prompt_tokens":890,"completion_tokens":56,"total_tokens":946}}
+```
--- a/plugins/wasm-go/extensions/ai-agent/README_EN.md
+++ b/plugins/wasm-go/extensions/ai-agent/README_EN.md
@@ -4,7 +4,7 @@ keywords: [ AI Gateway, AI Agent ]
 description: AI Agent plugin configuration reference
 ---
 ## Functional Description
-A customizable API AI Agent that supports configuring HTTP method types as GET and POST APIs. Supports multiple dialogue rounds, streaming and non-streaming modes.  
+A customizable API AI Agent that supports configuring HTTP method types as GET and POST APIs. Supports multiple dialogue rounds, streaming and non-streaming modes, support for formatting results as custom json.  
 The agent flow chart is as follows:  
 ![ai-agent](https://github.com/user-attachments/assets/b0761a0c-1afa-496c-a98e-bb9f38b340f8)  

@@ -20,6 +20,7 @@ Plugin execution priority: `200`
 | `llm`            | object    | Required    | -             | Configuration information for AI service provider  |
 | `apis`           | object    | Required    | -             | Configuration information for external API service provider  |
 | `promptTemplate` | object    | Optional    | -             | Configuration information for Agent ReAct template  |
+| `jsonResp`       | object    | Optional    | -             | Configuring json formatting information  |

 The configuration fields for `llm` are as follows:  
 | Name               | Data Type | Requirement | Default Value | Description                         |
@@ -71,7 +72,13 @@ The configuration fields for `chTemplate` and `enTemplate` are as follows:
 | `observation`   | string    | Optional    | -             | The observation part of the Agent ReAct template     |
 | `thought2`      | string    | Optional    | -             | The thought2 part of the Agent ReAct template       |

-## Usage Example
+The configuration fields for `jsonResp` are as follows:  
+| Name               | Data Type | Requirement | Default Value | Description                         |
+|--------------------|-----------|-------------|---------------|------------------------------------|
+| `enable`           | bool      | Optional    | -             | Whether to enable json formatting.  |
+| `jsonSchema`       | string    | Optional    | -             | Custom json schema               |
+
+## Usage Example-disable json formatting
 **Configuration Information**  
 ```yaml  
 llm:  
@@ -335,3 +342,68 @@ curl 'http://<replace with gateway public IP>/api/openai/v1/chat/completions' \
 {"id":"65dcf12c-61ff-9e68-bffa-44fc9e6070d5","choices":[{"index":0,"message":{"role":"assistant","content":" The German translation of \"Hail Hydra!\" is \"Hoch lebe Hydra!\"."},"finish_reason":"stop"}],"created":1724043865,"model":"qwen-max-0403","object":"chat.completion","usage":{"prompt_tokens":908,"completion_tokens":52,"total_tokens":960}}  
 ```  

+## Usage Example-enable json formatting
+**Configuration Information**  
+Add jsonResp configuration to the above configuration
+```yaml
+jsonResp:
+  enable: true
+```
+
+**Request Example**  
+```shell
+curl 'http://<replace with gateway public IP>/api/openai/v1/chat/completions' \
+-H 'Accept: application/json, text/event-stream' \
+-H 'Content-Type: application/json' \
+--data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role":"user","content":"What is the current weather in Beijing ?"}],"presence_penalty":0,"temperature":0,"top_p":0}'
+```
+
+**Response Example**
+
+```json
+{"id":"ebd6ea91-8e38-9e14-9a5b-90178d2edea4","choices":[{"index":0,"message":{"role":"assistant","content": "{\"city\": \"BeiJing\", \"weather_condition\": \"cloudy\", \"temperature\": \"19℃\", \"data_update_time\": \"Oct 9, 2024, at 16:37\"}"},"finish_reason":"stop"}],"created":1723187991,"model":"qwen-max-0403","object":"chat.completion","usage":{"prompt_tokens":890,"completion_tokens":56,"total_tokens":946}}
+```
+If you don't customise the json schema, the big model will automatically generate a json format
+
+**Configuration Information**
+Add custom json schema configuration
+```yaml
+jsonResp:
+  enable: true
+  jsonSchema:
+    title: WeatherSchema
+    type: object
+    properties:
+      location:
+        type: string
+        description: city name.
+      weather:
+        type: string
+        description: weather conditions.
+      temperature:
+        type: string
+        description: temperature.
+      update_time:
+        type: string
+        description: the update time of data.
+    required:
+      - location
+      - weather
+      - temperature
+    additionalProperties: false
+```
+
+**Request Example**
+
+```shell
+curl 'http://<replace with gateway public IP>/api/openai/v1/chat/completions' \
+-H 'Accept: application/json, text/event-stream' \
+-H 'Content-Type: application/json' \
+--data-raw '{"model":"qwen","frequency_penalty":0,"max_tokens":800,"stream":false,"messages":[{"role":"user","content":"What is the current weather in Beijing ?"}],"presence_penalty":0,"temperature":0,"top_p":0}'
+```
+
+**Response Example**
+
+```json
+{"id":"ebd6ea91-8e38-9e14-9a5b-90178d2edea4","choices":[{"index":0,"message":{"role":"assistant","content": "{\"location\": \"Beijing\", \"weather\": \"cloudy\", \"temperature\": \"19℃\", \"update_time\": \"Oct 9, 2024, at 16:37\"}"},"finish_reason":"stop"}],"created":1723187991,"model":"qwen-max-0403","object":"chat.completion","usage":{"prompt_tokens":890,"completion_tokens":56,"total_tokens":946}}
+```
--- a/plugins/wasm-go/extensions/ai-agent/config.go
+++ b/plugins/wasm-go/extensions/ai-agent/config.go
@@ -211,6 +211,15 @@ type LLMInfo struct {
 	MaxTokens int64 `yaml:"maxToken" json:"maxTokens"`
 }

+type JsonResp struct {
+	// @Title zh-CN Enable
+	// @Description zh-CN 是否要启用json格式化输出
+	Enable bool `yaml:"enable" json:"enable"`
+	// @Title zh-CN Json Schema
+	// @Description zh-CN 用以验证响应json的Json Schema, 为空则只验证返回的响应是否为合法json
+	JsonSchema map[string]interface{} `required:"false" json:"jsonSchema" yaml:"jsonSchema"`
+}
+
 type PluginConfig struct {
 	// @Title zh-CN 返回 HTTP 响应的模版
 	// @Description zh-CN 用 %s 标记需要被 cache value 替换的部分
@@ -225,6 +234,7 @@ type PluginConfig struct {
 	LLMClient      wrapper.HttpClient `yaml:"-" json:"-"`
 	APIsParam      []APIsParam        `yaml:"-" json:"-"`
 	PromptTemplate PromptTemplate     `yaml:"promptTemplate" json:"promptTemplate"`
+	JsonResp       JsonResp           `yaml:"jsonResp" json:"jsonResp"`
 }

 func initResponsePromptTpl(gjson gjson.Result, c *PluginConfig) {
@@ -402,3 +412,15 @@ func initLLMClient(gjson gjson.Result, c *PluginConfig) {
 		Host: c.LLMInfo.Domain,
 	})
 }
+
+func initJsonResp(gjson gjson.Result, c *PluginConfig) {
+	c.JsonResp.Enable = false
+	if c.JsonResp.Enable = gjson.Get("jsonResp.enable").Bool(); c.JsonResp.Enable {
+		c.JsonResp.JsonSchema = nil
+		if jsonSchemaValue := gjson.Get("jsonResp.jsonSchema"); jsonSchemaValue.Exists() {
+			if schemaValue, ok := jsonSchemaValue.Value().(map[string]interface{}); ok {
+				c.JsonResp.JsonSchema = schemaValue
+			}
+		}
+	}
+}
--- a/plugins/wasm-go/extensions/ai-agent/main.go
+++ b/plugins/wasm-go/extensions/ai-agent/main.go
@@ -2,8 +2,10 @@ package main

 import (
 	"encoding/json"
+	"errors"
 	"fmt"
 	"net/http"
+	"net/url"
 	"regexp"
 	"strings"

@@ -47,6 +49,8 @@ func parseConfig(gjson gjson.Result, c *PluginConfig, log wrapper.Log) error {

 	initLLMClient(gjson, c)

+	initJsonResp(gjson, c)
+
 	return nil
 }

@@ -76,10 +80,10 @@ func firstReq(ctx wrapper.HttpContext, config PluginConfig, prompt string, rawRe
 		log.Debugf("[onHttpRequestBody] newRequestBody: %s", string(newbody))
 		err := proxywasm.ReplaceHttpRequestBody(newbody)
 		if err != nil {
-			log.Debug("替换失败")
+			log.Debugf("failed replace err: %s", err.Error())
 			proxywasm.SendHttpResponse(200, [][2]string{{"content-type", "application/json; charset=utf-8"}}, []byte(fmt.Sprintf(config.ReturnResponseTemplate, "替换失败"+err.Error())), -1)
 		}
-		log.Debug("[onHttpRequestBody] request替换成功")
+		log.Debug("[onHttpRequestBody] replace request success")
 		return types.ActionContinue
 	}
 }
@@ -175,11 +179,103 @@ func onHttpResponseHeaders(ctx wrapper.HttpContext, config PluginConfig, log wra
 	return types.ActionContinue
 }

-func toolsCallResult(ctx wrapper.HttpContext, config PluginConfig, content string, rawResponse Response, log wrapper.Log, statusCode int, responseBody []byte) {
+func extractJson(bodyStr string) (string, error) {
+	// simply extract json from response body string
+	startIndex := strings.Index(bodyStr, "{")
+	endIndex := strings.LastIndex(bodyStr, "}") + 1
+
+	// if not found
+	if startIndex == -1 || startIndex >= endIndex {
+		return "", errors.New("cannot find json in the response body")
+	}
+
+	jsonStr := bodyStr[startIndex:endIndex]
+
+	// attempt to parse the JSON
+	var result map[string]interface{}
+	err := json.Unmarshal([]byte(jsonStr), &result)
+	if err != nil {
+		return "", err
+	}
+	return jsonStr, nil
+}
+
+func jsonFormat(llmClient wrapper.HttpClient, llmInfo LLMInfo, jsonSchema map[string]interface{}, assistantMessage Message, actionInput string, headers [][2]string, streamMode bool, rawResponse Response, log wrapper.Log) string {
+	prompt := fmt.Sprintf(prompttpl.Json_Resp_Template, jsonSchema, actionInput)
+
+	messages := []dashscope.Message{{Role: "user", Content: prompt}}
+
+	completion := dashscope.Completion{
+		Model:    llmInfo.Model,
+		Messages: messages,
+	}
+
+	completionSerialized, _ := json.Marshal(completion)
+	var content string
+	err := llmClient.Post(
+		llmInfo.Path,
+		headers,
+		completionSerialized,
+		func(statusCode int, responseHeaders http.Header, responseBody []byte) {
+			//得到gpt的返回结果
+			var responseCompletion dashscope.CompletionResponse
+			_ = json.Unmarshal(responseBody, &responseCompletion)
+			log.Infof("[jsonFormat] content: %s", responseCompletion.Choices[0].Message.Content)
+			content = responseCompletion.Choices[0].Message.Content
+			jsonStr, err := extractJson(content)
+			if err != nil {
+				log.Debugf("[onHttpRequestBody] extractJson err: %s", err.Error())
+				jsonStr = content
+			}
+
+			if streamMode {
+				stream(jsonStr, rawResponse, log)
+			} else {
+				noneStream(assistantMessage, jsonStr, rawResponse, log)
+			}
+		}, uint32(llmInfo.MaxExecutionTime))
+	if err != nil {
+		log.Debugf("[onHttpRequestBody] completion err: %s", err.Error())
+		proxywasm.ResumeHttpRequest()
+	}
+	return content
+}
+
+func noneStream(assistantMessage Message, actionInput string, rawResponse Response, log wrapper.Log) {
+	assistantMessage.Role = "assistant"
+	assistantMessage.Content = actionInput
+	rawResponse.Choices[0].Message = assistantMessage
+	newbody, err := json.Marshal(rawResponse)
+	if err != nil {
+		proxywasm.ResumeHttpResponse()
+		return
+	} else {
+		proxywasm.ReplaceHttpResponseBody(newbody)
+
+		log.Debug("[onHttpResponseBody] replace response success")
+		proxywasm.ResumeHttpResponse()
+	}
+}
+
+func stream(actionInput string, rawResponse Response, log wrapper.Log) {
+	headers := [][2]string{{"content-type", "text/event-stream; charset=utf-8"}}
+	proxywasm.ReplaceHttpResponseHeaders(headers)
+	// Remove quotes from actionInput
+	actionInput = strings.Trim(actionInput, "\"")
+	returnStreamResponseTemplate := `data:{"id":"%s","choices":[{"index":0,"delta":{"role":"assistant","content":"%s"},"finish_reason":"stop"}],"model":"%s","object":"chat.completion","usage":{"prompt_tokens":%d,"completion_tokens":%d,"total_tokens":%d}}` + "\n\ndata:[DONE]\n\n"
+	newbody := fmt.Sprintf(returnStreamResponseTemplate, rawResponse.ID, actionInput, rawResponse.Model, rawResponse.Usage.PromptTokens, rawResponse.Usage.CompletionTokens, rawResponse.Usage.TotalTokens)
+	log.Infof("[onHttpResponseBody] newResponseBody: ", newbody)
+	proxywasm.ReplaceHttpResponseBody([]byte(newbody))
+
+	log.Debug("[onHttpResponseBody] replace response success")
+	proxywasm.ResumeHttpResponse()
+}
+
+func toolsCallResult(ctx wrapper.HttpContext, llmClient wrapper.HttpClient, llmInfo LLMInfo, jsonResp JsonResp, aPIsParam []APIsParam, aPIClient []wrapper.HttpClient, content string, rawResponse Response, log wrapper.Log, statusCode int, responseBody []byte) {
 	if statusCode != http.StatusOK {
 		log.Debugf("statusCode: %d", statusCode)
 	}
-	log.Info("========函数返回结果========")
+	log.Info("========function result========")
 	log.Infof(string(responseBody))

 	observation := "Observation: " + string(responseBody)
@@ -187,15 +283,15 @@ func toolsCallResult(ctx wrapper.HttpContext, config PluginConfig, content strin
 	dashscope.MessageStore.AddForUser(observation)

 	completion := dashscope.Completion{
-		Model:     config.LLMInfo.Model,
+		Model:     llmInfo.Model,
 		Messages:  dashscope.MessageStore,
-		MaxTokens: config.LLMInfo.MaxTokens,
+		MaxTokens: llmInfo.MaxTokens,
 	}

-	headers := [][2]string{{"Content-Type", "application/json"}, {"Authorization", "Bearer " + config.LLMInfo.APIKey}}
+	headers := [][2]string{{"Content-Type", "application/json"}, {"Authorization", "Bearer " + llmInfo.APIKey}}
 	completionSerialized, _ := json.Marshal(completion)
-	err := config.LLMClient.Post(
-		config.LLMInfo.Path,
+	err := llmClient.Post(
+		llmInfo.Path,
 		headers,
 		completionSerialized,
 		func(statusCode int, responseHeaders http.Header, responseBody []byte) {
@@ -205,42 +301,31 @@ func toolsCallResult(ctx wrapper.HttpContext, config PluginConfig, content strin
 			log.Infof("[toolsCall] content: %s", responseCompletion.Choices[0].Message.Content)

 			if responseCompletion.Choices[0].Message.Content != "" {
-				retType, actionInput := toolsCall(ctx, config, responseCompletion.Choices[0].Message.Content, rawResponse, log)
+				retType, actionInput := toolsCall(ctx, llmClient, llmInfo, jsonResp, aPIsParam, aPIClient, responseCompletion.Choices[0].Message.Content, rawResponse, log)
 				if retType == types.ActionContinue {
 					//得到了Final Answer
 					var assistantMessage Message
+					var streamMode bool
 					if ctx.GetContext(StreamContextKey) == nil {
-						assistantMessage.Role = "assistant"
-						assistantMessage.Content = actionInput
-						rawResponse.Choices[0].Message = assistantMessage
-						newbody, err := json.Marshal(rawResponse)
-						if err != nil {
-							proxywasm.ResumeHttpResponse()
-							return
+						streamMode = false
+						if jsonResp.Enable {
+							jsonFormat(llmClient, llmInfo, jsonResp.JsonSchema, assistantMessage, actionInput, headers, streamMode, rawResponse, log)
 						} else {
-							proxywasm.ReplaceHttpResponseBody(newbody)
-
-							log.Debug("[onHttpResponseBody] response替换成功")
-							proxywasm.ResumeHttpResponse()
+							noneStream(assistantMessage, actionInput, rawResponse, log)
 						}
 					} else {
-						headers := [][2]string{{"content-type", "text/event-stream; charset=utf-8"}}
-						proxywasm.ReplaceHttpResponseHeaders(headers)
-						// Remove quotes from actionInput
-						actionInput = strings.Trim(actionInput, "\"")
-						returnStreamResponseTemplate := `data:{"id":"%s","choices":[{"index":0,"delta":{"role":"assistant","content":"%s"},"finish_reason":"stop"}],"model":"%s","object":"chat.completion","usage":{"prompt_tokens":%d,"completion_tokens":%d,"total_tokens":%d}}` + "\n\ndata:[DONE]\n\n"
-						newbody := fmt.Sprintf(returnStreamResponseTemplate, rawResponse.ID, actionInput, rawResponse.Model, rawResponse.Usage.PromptTokens, rawResponse.Usage.CompletionTokens, rawResponse.Usage.TotalTokens)
-						log.Infof("[onHttpResponseBody] newResponseBody: ", newbody)
-						proxywasm.ReplaceHttpResponseBody([]byte(newbody))
-
-						log.Debug("[onHttpResponseBody] response替换成功")
-						proxywasm.ResumeHttpResponse()
+						streamMode = true
+						if jsonResp.Enable {
+							jsonFormat(llmClient, llmInfo, jsonResp.JsonSchema, assistantMessage, actionInput, headers, streamMode, rawResponse, log)
+						} else {
+							stream(actionInput, rawResponse, log)
+						}
 					}
 				}
 			} else {
 				proxywasm.ResumeHttpRequest()
 			}
-		}, uint32(config.LLMInfo.MaxExecutionTime))
+		}, uint32(llmInfo.MaxExecutionTime))
 	if err != nil {
 		log.Debugf("[onHttpRequestBody] completion err: %s", err.Error())
 		proxywasm.ResumeHttpRequest()
@@ -294,7 +379,7 @@ func outputParser(response string, log wrapper.Log) (string, string) {
 	return "", ""
 }

-func toolsCall(ctx wrapper.HttpContext, config PluginConfig, content string, rawResponse Response, log wrapper.Log) (types.Action, string) {
+func toolsCall(ctx wrapper.HttpContext, llmClient wrapper.HttpClient, llmInfo LLMInfo, jsonResp JsonResp, aPIsParam []APIsParam, aPIClient []wrapper.HttpClient, content string, rawResponse Response, log wrapper.Log) (types.Action, string) {
 	dashscope.MessageStore.AddForAssistant(content)

 	action, actionInput := outputParser(content, log)
@@ -305,9 +390,9 @@ func toolsCall(ctx wrapper.HttpContext, config PluginConfig, content string, raw
 	}
 	count := ctx.GetContext(ToolCallsCount).(int)
 	count++
-	log.Debugf("toolCallsCount:%d, config.LLMInfo.MaxIterations=%d", count, config.LLMInfo.MaxIterations)
+	log.Debugf("toolCallsCount:%d, config.LLMInfo.MaxIterations=%d", count, llmInfo.MaxIterations)
 	//函数递归调用次数，达到了预设的循环次数，强制结束
-	if int64(count) > config.LLMInfo.MaxIterations {
+	if int64(count) > llmInfo.MaxIterations {
 		ctx.SetContext(ToolCallsCount, 0)
 		return types.ActionContinue, ""
 	} else {
@@ -316,15 +401,14 @@ func toolsCall(ctx wrapper.HttpContext, config PluginConfig, content string, raw

 	//没得到最终答案

-	var url string
+	var urlStr string
 	var headers [][2]string
 	var apiClient wrapper.HttpClient
 	var method string
 	var reqBody []byte
-	var key string
 	var maxExecutionTime int64

-	for i, apisParam := range config.APIsParam {
+	for i, apisParam := range aPIsParam {
 		maxExecutionTime = apisParam.MaxExecutionTime
 		for _, tools_param := range apisParam.ToolsParam {
 			if action == tools_param.ToolName {
@@ -340,28 +424,37 @@ func toolsCall(ctx wrapper.HttpContext, config PluginConfig, content string, raw

 				method = tools_param.Method

-				// 组装 headers 和 key
-				headers = [][2]string{{"Content-Type", "application/json"}}
-				if apisParam.APIKey.Name != "" {
-					if apisParam.APIKey.In == "query" {
-						key = "?" + apisParam.APIKey.Name + "=" + apisParam.APIKey.Value
-					} else if apisParam.APIKey.In == "header" {
-						headers = append(headers, [2]string{"Authorization", apisParam.APIKey.Name + " " + apisParam.APIKey.Value})
+				// 组装 URL 和请求体
+				urlStr = apisParam.URL + tools_param.Path
+
+				// 解析URL模板以查找路径参数
+				urlParts := strings.Split(urlStr, "/")
+				for i, part := range urlParts {
+					if strings.Contains(part, "{") && strings.Contains(part, "}") {
+						for _, param := range tools_param.ParamName {
+							paramNameInPath := part[1 : len(part)-1]
+							if paramNameInPath == param {
+								if value, ok := data[param]; ok {
+									// 删除已经使用过的
+									delete(data, param)
+									// 替换模板中的占位符
+									urlParts[i] = url.QueryEscape(value.(string))
+								}
+							}
+						}
 					}
 				}

-				// 组装 URL 和请求体
-				url = apisParam.URL + tools_param.Path + key
+				// 重新组合URL
+				urlStr = strings.Join(urlParts, "/")
+
+				queryParams := make([][2]string, 0)
 				if method == "GET" {
-					queryParams := make([]string, 0, len(tools_param.ParamName))
 					for _, param := range tools_param.ParamName {
 						if value, ok := data[param]; ok {
-							queryParams = append(queryParams, fmt.Sprintf("%s=%v", param, value))
+							queryParams = append(queryParams, [2]string{param, fmt.Sprintf("%v", value)})
 						}
 					}
-					if len(queryParams) > 0 {
-						url += "&" + strings.Join(queryParams, "&")
-					}
 				} else if method == "POST" {
 					var err error
 					reqBody, err = json.Marshal(data)
@@ -371,9 +464,30 @@ func toolsCall(ctx wrapper.HttpContext, config PluginConfig, content string, raw
 					}
 				}

-				log.Infof("url: %s", url)
+				// 组装 headers 和 key
+				headers = [][2]string{{"Content-Type", "application/json"}}
+				if apisParam.APIKey.Name != "" {
+					if apisParam.APIKey.In == "query" {
+						queryParams = append(queryParams, [2]string{apisParam.APIKey.Name, apisParam.APIKey.Value})
+					} else if apisParam.APIKey.In == "header" {
+						headers = append(headers, [2]string{"Authorization", apisParam.APIKey.Name + " " + apisParam.APIKey.Value})
+					}
+				}

-				apiClient = config.APIClient[i]
+				if len(queryParams) > 0 {
+					// 将 key 拼接到 url 后面
+					urlStr += "?"
+					for i, param := range queryParams {
+						if i != 0 {
+							urlStr += "&"
+						}
+						urlStr += url.QueryEscape(param[0]) + "=" + url.QueryEscape(param[1])
+					}
+				}
+
+				log.Debugf("url: %s", urlStr)
+
+				apiClient = aPIClient[i]
 				break
 			}
 		}
@@ -382,11 +496,11 @@ func toolsCall(ctx wrapper.HttpContext, config PluginConfig, content string, raw
 	if apiClient != nil {
 		err := apiClient.Call(
 			method,
-			url,
+			urlStr,
 			headers,
 			reqBody,
 			func(statusCode int, responseHeaders http.Header, responseBody []byte) {
-				toolsCallResult(ctx, config, content, rawResponse, log, statusCode, responseBody)
+				toolsCallResult(ctx, llmClient, llmInfo, jsonResp, aPIsParam, aPIClient, content, rawResponse, log, statusCode, responseBody)
 			}, uint32(maxExecutionTime))
 		if err != nil {
 			log.Debugf("tool calls error: %s", err.Error())
@@ -415,7 +529,7 @@ func onHttpResponseBody(ctx wrapper.HttpContext, config PluginConfig, body []byt
 	//如果gpt返回的内容不是空的
 	if rawResponse.Choices[0].Message.Content != "" {
 		//进入agent的循环思考，工具调用的过程中
-		retType, _ := toolsCall(ctx, config, rawResponse.Choices[0].Message.Content, rawResponse, log)
+		retType, _ := toolsCall(ctx, config.LLMClient, config.LLMInfo, config.JsonResp, config.APIsParam, config.APIClient, rawResponse.Choices[0].Message.Content, rawResponse, log)
 		return retType
 	} else {
 		return types.ActionContinue
--- a/plugins/wasm-go/extensions/ai-agent/promptTpl/prompt.go
+++ b/plugins/wasm-go/extensions/ai-agent/promptTpl/prompt.go
@@ -167,3 +167,7 @@ Action:` + "```" + `
 %s
 Question: %s
 `
+const Json_Resp_Template = `
+Given the Json Schema: %s, please help me convert the following content to a pure json: %s
+Do not respond other content except the pure json!!!!
+`