Requirements for the AIPluginTool to work with Wolfram?

[qamar-gtschool]

Hi! So I've been trying to use the AIPluginTool to essentially recreate success with a prompt that currently works with the retail version of ChatGPT + Wolfram Plugin. I've not been having much luck with it. Here's an outline of my code and the output.

The only response I get is the API usage guide. Could this be an input issue, or is there a specific account-level requirement I need, like getting pro and API access?

I've searched Twitter and the threads here, but it doesn't seem like many people are discussing the plugin tool. Any insights would be appreciated.

[dosubot[bot]]

🤖

Hello,

Based on the information you've provided, it seems like you're having trouble with the AIPluginTool in the LangChain framework when trying to use it with the Wolfram Plugin.

The AIPluginTool requires an AIPlugin instance to work, which is created from a URL that points to the plugin's JSON configuration. This configuration should include several fields such as schema_version, name_for_model, name_for_human, description_for_model, description_for_human, auth, api, logo_url, contact_email, and legal_info_url. You can find more details about these fields in the context provided above.

The AIPluginTool does not require any specific account-level access like pro or API, as it only fetches the OpenAPI spec and does not perform any operations that would require authentication. However, the Wolfram Alpha plugin specifically requires a developer account and an APP ID. This APP ID should be saved in the WOLFRAM_ALPHA_APPID environment variable. Additionally, the wolframalpha Python package needs to be installed.

The issues you're experiencing could be due to an input problem if the URL you're providing does not point to a valid JSON configuration for the plugin. Make sure the URL is correct and the JSON configuration includes all the required fields. Also, ensure that you have the necessary Wolfram Alpha developer account, APP ID, and the wolframalpha Python package installed.

The Wolfram Alpha plugin has two endpoints: /api/v1/cloud-plugin and /api/v1/llm-api. Both endpoints require an 'input' parameter in the query string. If this parameter is missing, you will receive a 400 error. Unauthorized access will result in a 403 error, and too many requests will result in a 503 error. Make sure your requests to these endpoints include the required 'input' parameter and that you are not exceeding the request limit.

You can find more details about the AIPluginTool and the Wolfram Alpha plugin in the LangChain repository at these locations:

• libs/langchain/langchain/tools/plugin.py
• libs/langchain/langchain/utilities/wolfram_alpha.py
• libs/langchain/tests/unit_tests/examples/test_specs/wolframalpha/apispec.json

I hope this helps! If you have any more questions or need further clarification, feel free to ask.

Sources

• libs/langchain/langchain/tools/plugin.py
• libs/langchain/langchain/utilities/wolfram_alpha.py
• libs/langchain/tests/unit_tests/examples/test_specs/wolframalpha/apispec.json

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[qamar-gtschool]

@dosu-bot
I have the WOLFRAM_ALPHA_APPID environment variable set. One of the commands I can see LangChain attempt is to call getWolframAlphaResults with the following input: {"input": "Solve 4x + 2 = 12"}, which is what the API spec asks it to do. The response from the plugin is still a large body of text with the usage guide.

I am not currently using any of the endpoints you listed. I assumed it's automatically loaded with this line:
tool = AIPluginTool.from_plugin_url("https://www.wolframalpha.com/.well-known/ai-plugin.json"). Should I be loading the plugin tool differently, using the endpoints?

[dosubot[bot]]

🤖

Hello,

The issue you're experiencing might be due to the way the run method in the WolframAlphaAPIWrapper class is handling the results from the Wolfram Alpha API.

The run method is trying to extract the assumption and answer from the result pods. However, if the first pod does not contain an assumption or the first result does not contain an answer, it will raise a StopIteration exception and return "Wolfram Alpha wasn't able to answer it".

If the answer is None or an empty string, it will return "No good Wolfram Alpha Result was found".

This could be the reason why you're not getting the expected results. If the first pod or result does not contain the expected data, you will not get the expected output.

You might want to modify the run method to better handle the results from the Wolfram Alpha API. For example, you could iterate over all the pods and results until you find a valid assumption and answer.

Here's an example of how you could modify the run method:

def run(self, query: str) -> str:
    """Run query through WolframAlpha and parse result."""
    res = self.wolfram_client.query(query)

    assumption = None
    answer = None

    for pod in res.pods:
        if assumption is None and pod.text is not None and pod.text != "":
            assumption = pod.text

    for result in res.results:
        if answer is None and result.text is not None and result.text != "":
            answer = result.text

    if assumption is None or answer is None:
        return "Wolfram Alpha wasn't able to answer it"
    else:
        return f"Assumption: {assumption} \nAnswer: {answer}"

This version of the run method will iterate over all the pods and results until it finds a valid assumption and answer. If it can't find a valid assumption or answer, it will return "Wolfram Alpha wasn't able to answer it".

In the context shared, the WolframAlphaQueryRun class in the tool.py file uses the run method from the WolframAlphaAPIWrapper class. If you modify the run method as suggested, it should resolve the issue you're experiencing.

Please let me know if this helps or if you need further assistance.

Sources

• libs/langchain/langchain/utilities/wolfram_alpha.py
• libs/langchain/langchain/tools/wolfram_alpha/tool.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[qamar-gtschool]

@dosu-beta

Hi. Your response assumes I'm using the WolframAlphaAPIWrapper class directly. I am using the AIPluginTool, and I am loading the Wolfram plugin via this URL: https://www.wolframalpha.com/.well-known/ai-plugin.json.

I need the Wolfram plugin specifically, so I can use it the same way I use it on the web form. Are there any resources or experiences from anywhere or anyone else using the AIPluginTool to call Wolfram? My input seems to be correct, but the response is always a long string of text with the API Usage Guide.

[dosubot[bot]]

🤖

Hello,

From your description, it seems like you're doing everything correctly. The AIPluginTool.from_plugin_url method should indeed load the plugin from the provided URL, and the getWolframAlphaResults method should be able to handle the input you're providing.

However, the fact that you're receiving a long string of text with the API Usage Guide instead of the expected results suggests that there might be an issue with the way the Wolfram Alpha API is being called or the response is being processed.

Here are a few things you could check:

1. API Key: Make sure that the WOLFRAM_ALPHA_APPID environment variable is correctly set to your Wolfram Alpha API key. If the key is incorrect or missing, the API might return an error message or usage guide instead of the expected results.

2. API Call: Check the exact API call that is being made. You can do this by adding some debug logging to the WolframAlphaAPIWrapper class in the langchain.utilities.wolfram_alpha module. This will allow you to see the exact URL and parameters that are being used for the API call.

3. API Response: Similarly, you could add some logging to the WolframAlphaQueryRun class in the langchain.tools.wolfram_alpha module to see the exact response that is being returned by the API. This might give you some clues as to why you're getting the API Usage Guide instead of the expected results.

4. API Specification: It's possible that the OpenAPI specification returned by the https://www.wolframalpha.com/.well-known/ai-plugin.json URL is incorrect or outdated. You could try to manually download and inspect this specification to see if it matches the expected format and contains the correct information for the getWolframAlphaResults operation.

If none of these suggestions help, it would be useful to have more information about the exact error message or unexpected output you're getting. Could you provide the full text of the API Usage Guide that you're receiving, or any other error messages or logs that might be relevant?

I hope this helps, and please let me know if you have any further questions or if there's anything else I can do to assist you.

Sources

• libs/langchain/langchain/tools/plugin.py
• libs/langchain/langchain/utilities/wolfram_alpha.py
• libs/langchain/tests/integration_tests/utilities/test_wolfram_alpha_api.py
• libs/langchain/langchain/tools/wolfram_alpha/tool.py

---

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.