nexent bug [Bug]

[ganzhipeng111]

Nexent Version

No response

Problem Description

The agent created by Nexent cannot answer questions. The agent created by Nexent cannot answer questions, and it seems that there is a bug in this platform.

Reproduction Steps

No response

Additional Information

No response

[SimengBian]

It seems to be an error with tools. Did you use third-party MCP tools? Did other agents with no MCP tools work well? Please provide more info to locate this error. For example you can provide the exported json file of this agent.

[ganzhipeng111]

It seems to be an error with tools. Did you use third-party MCP tools? Did other agents with no MCP tools work well? Please provide more info to locate this error. For example you can provide the exported json file of this agent.

The agent I created does not use the MCP tool; it only uses the local knowledge base. However, the error shown in the above image still occurs when calling the local knowledge base.

medical_diagnosis_assistant_config.json

[ganzhipeng111]

回忆已完成
正在连接MCP服务器...
正在思考中...
思考：用户询问"什么是HIV"，这是一个关于艾滋病病毒的基础知识问题。根据我的专业知识，HIV是人类免疫缺陷病毒的缩写，是导致艾滋病的病毒。不过为了确保信息的准确性和全面性，我决定使用知识库搜索工具来获取更详细的专业信息。

工具调用中...
知识库检索中...
HIV 人类免疫缺陷病毒 艾滋病
阅读检索结果
病理知识.docx
💥 运行出错： Error in generating model output: Error code: 400 - {'error': {'message': 'Failed to deserialize the JSON body into the target type: messages[2]: invalid type: sequence, expected a string at line 1 column 10648', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}
💥 运行出错： Error in generating model output: Error code: 400 - {'error': {'message': 'Failed to deserialize the JSON body into the target type: messages[2]: invalid type: sequence, expected a string at line 1 column 10648', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}
💥 运行出错： Error in generating model output: Error code: 400 - {'error': {'message': 'Failed to deserialize the JSON body into the target type: messages[2]: invalid type: sequence, expected a string at line 1 column 10648', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}
💥 运行出错： Error in generating model output: Error code: 400 - {'error': {'message': 'Failed to deserialize the JSON body into the target type: messages[2]: invalid type: sequence, expected a string at line 1 column 10648', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}
💥 运行出错： Error in generating model output: Error code: 400 - {'error': {'message': 'Failed to deserialize the JSON body into the target type: messages[2]: invalid type: sequence, expected a string at line 1 column 10648', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_request_error'}}

[SimengBian]

Could you please take a screen shot of your knowledge base page? Also you can provide the files you use so that we can reproduce this error.

[ganzhipeng111]

Could you please take a screen shot of your knowledge base page? Also you can provide the files you use so that we can reproduce this error.

OK.Please test the file I sent.

甲状腺钙化对病理诊断的影响及时间梯度脱钙法的应用价值_刘诗文.pdf

病理知识.docx

[SimengBian]

After a series of investigations, we finally discovered the problem: You are using the deepseek-chat model provided by the official deepseek provider. This model does not currently support multimodal dialogue formats, while Nexent is designed primarily for multimodal communication. You can try using the deepseek model provided by siliconflow provider. They have encapsulated the dialogue interface and can support multimodal dialogue. Specifically, the interface we use is in the form of:
{ "role":"user", "content":[ { "type":"text", "text":"prompt" } ] }
While deepseek only accepts formats such as:
{ "role":"user", "content":"prompt" }

Official Reference: cline/cline#230

[SimengBian]

We will include this information in the notes of the documentation. You can use other models to continue your development. Thank you!

[ganzhipeng111]

We will include this information in the notes of the documentation. You can use other models to continue your development. Thank you!

OK,thank you very much！