Does the value attribute of the @Tool annotation support variables like {{varName}}?

[growmuye]

@Tool(name = "queryFreeMeetingRooms", value = "Today is {{time}}, and the user wants to inquire about available meeting rooms.")

[langchain4j]

Hi @growmuye no, this is not supported (yet). I am not sure information about current date/time should be associated with specific tool, could you please elaborate? You can automatically provide current date/time in the system/user message, see example here: https://github.com/langchain4j/langchain4j-examples/blob/aae3b1356057f10a6d2a12f618f7bc673dd7f35e/spring-boot-example/src/main/java/dev/example/CustomerSupportAgent.java#L11
More info: https://github.com/langchain4j/langchain4j/blob/38049a197bac084b9f980a7edaf9619e0115403d/langchain4j-core/src/main/java/dev/langchain4j/model/input/PromptTemplate.java#L18C5-L18C5

[growmuye]

Thank you for your response. I have encountered an issue where I need LLM to be aware of the current time for some date calculations, so that LLM can perform accurate calculations. Initially, I planned to include the current time in the function description, but now I have declared the current time in the system and it is also effective.

[growmuye]

For the past three weeks, I have been using Langchain4j for development, and the biggest issue I've encountered is configurability. For example, I need to manage, distribute, respond, and rollback the system messages, function information, and temperature, among others, in conjunction with our company's configuration center. I was wondering if you have any suggestions or recommendations regarding configurability?

[growmuye]

For example, annotations like Jackson can be configured on a class to allow custom logic implementation within the class.

[growmuye]

I have a feeling about configuration: splitting a complete configuration into multiple pieces and managing them in different places can be quite cumbersome. I have thought about doing it in the following way:

1. Provide a complete protocol based on the LLM service provider specification.
2. Then use AiService.load(config).bind(“reserve_meetingroom_by_meetingroomname”, an object to execute the function logic).I haven’t figured out the details yet, but I will give you feedback once I have a more complete idea.

Here is an example of a protocol based on the LLM service provider specification:

    "model":"gpt-35-turbo-16k",
    "function_call":"auto",
    "messages": [
        ...message history
    ],
    "functions": [{
            "name": "reserve_meetingroom_by_meetingroomname",
            "description": "如果知道用户预订会议室的名称，可使用该函数；",
            "parameters": {
                "type": "object",
                "properties": {
                    "meeting_roomname": {
                        "type": "string",
                        "description": "会议室名称"
                    },
                    "meeting_starttime": {
                        "type": "string",
                        "description": "开始时间"
                    },
                    "meeting_endtime": {
                        "type": "string",
                        "description": "结束时间"
                    }
                },
                "required": ["meeting_roomname","meeting_starttime","meeting_endtime"]
            }
        },
        {
            "name": "reserve_meetingroom",
            "description": "预定会议室",
            "parameters": {
                "type": "object",
                "properties": {
                    "meeting_starttime": {
                        "type": "string",
                        "description": "开始时间"
                    },
                    "meeting_endtime": {
                        "type": "string",
                        "description": "结束时间"
                    },
                    "number_of_attendees ": {
                        "type": "string",
                        "description": "人数"
                    },
                    "need_projector": {
                        "type": "string",
                        "description": "电视",
                        "enum": ["true", "false", "null"]
                    }
                },
                "required": ["meeting_starttime","meeting_endtime","number_of_attendees","need_projector"]
            }
        }
    ],
    "temperature":1,
    "stream":false
}```

[growmuye]

OK... I have one final suggestion. I would like to know if you could consider it:

Context is a pass-through object that I will use to store the data required for business purposes when initiating a conversation. The functions within will utilize this data to complete the business logic.