Need help with OpenAPI agent to call Internal/External API

[mohit19906]

I have a set of APIs. These APIs are internal to my organization. I want to use the OpenAPI agent to take the user input, and call the API based on the input. I have gone through Langchain document https://python.langchain.com/docs/integrations/toolkits/openapi.
Although I created a sample OpenAPI specification file to test. My specification file contains below content -

**openapi: 3.0.0
info:
title: OpenAI API
description: APIs for terminate member
version: 2.0.0
servers:

• url: http://127.0.0.1:5000
  tags:
• name: OpenAI
  description: The OpenAI REST API
  paths:
  "/api/terminateApi":
  post:
  operationId: terminateMember
  tags:
  - OpenAI
  summary: Terminates a member
  requestBody:
  required: true
  responses:
  '200':
  description: OK
  x-oaiMeta:
  name: terminate
  group: terminate
  path: terminate
  x-oaiMeta:
  groups:
  • id: terminate
    title: Terminate Member
    description: Used to terminate a member and returns 'Your tansaction is sucessfully completed' when the member
    is successfully terminated and returns 'invalid member' if member is not terminated.**

I am a newbie so having trouble understanding the documentation. I sifted through the internet and youtube but didn't find anything helpful.
I need help with below -

1. When I feed in the "can you terminate" , it tried to look for "Get" API even though the specification file mentions 'POST'. When I change the input to just "terminate", then it works fine by looking for 'Post' method. Why it keeps changing based on user input. How can I fix it.
2. How do I add the Request Body on the run time to call any API.
3. Where can I get better documentation of how to do these things.

[levalencia]

can you paste the code of how you setup the tools?

[mohit19906]

Thanks for your response.
I am triggering the checkAgent API with the Prompt as 'Can you terminate' from Postman. This endpoint has the logic to read OpenAPI specification file and call another API which is 'terminateApi'. terminate.yml file contains specification which I mentioned in my original question. Below is the code -

@app.route('/api/checkAgent', methods=['POST'])
def apiAgent():
    prompt = request.get_json(['prompt'])
    print(prompt["prompt"])
    with open("terminate.yml") as f:
        raw_openai_api_spec = yaml.load(f, Loader=yaml.Loader)
    openai_api_spec = reduce_openapi_spec(raw_openai_api_spec)
   
    requests_wrapper = RequestsWrapper(headers={'Content-Type':'application/json'})
    spotify_agent = planner.create_openapi_agent(openai_api_spec, requests_wrapper, get_openai_model())
    return spotify_agent.run(prompt['prompt'])

@app.route('/api/terminateApi',methods=['POST'])
def checkAPI():
    print("I am in check API ")
    return "Termination Successful"

[dosubot[bot]]

Answer generated by a 🤖

Answer

I understand that you're having a few issues with the OpenAPI agent in LangChain. Specifically, you're having trouble with the HTTP method selection based on user input, adding a request body at runtime, and finding comprehensive documentation. Let's address these issues one by one.

1. The OpenAPI agent in LangChain uses natural language processing to infer the HTTP method (GET, POST, etc.) from the user's input. For instance, if the input is "can you terminate", it might interpret this as a request for information, which typically corresponds to a GET request. However, if the input is simply "terminate", it interprets this as a command, which corresponds to a POST request. To have more control over this, you can specify the HTTP method when creating the APIOperation object. Here's how you can do it:

from langchain.tools import OpenAPISpec, APIOperation
spec = OpenAPISpec.from_url("your_specification_url")
operation = APIOperation.from_openapi_spec(spec, "/api/terminateApi", "post")

2. To add a request body at runtime, you can pass it as an argument when calling the chain. The request body should be a dictionary that matches the schema defined in your OpenAPI specification. Here's an example:

from langchain.chains import OpenAPIEndpointChain
from langchain.requests import Requests
from langchain.llms import OpenAI
llm = OpenAI(temperature=0, max_tokens=1000)
api_chain = OpenAPIEndpointChain.from_api_operation(
    operation,
    llm,
    requests=Requests(),
    verbose=False,
    return_intermediate_steps=True,  # Return request and response text
)
output = api_chain("terminate", request_body={"memberId": "123"})

3. As for documentation, you can find the official LangChain documentation at https://python.langchain.com/docs/. If you're finding it difficult to understand, it might be helpful to look at examples of how to use LangChain. If you're new to OpenAPI, the official OpenAPI Specification (https://swagger.io/specification/) might also be a good resource.

Remember, this is just one potential solution. If you find a better way to solve your issue or if you think this solution could be better explained in the documentation, feel free to make a pull request to update the documentation. This could help other users who might encounter the same issue in the future.

Sources

• docs/extras/modules/chains/additional/openapi.ipynb
• docs/extras/guides/evaluation/openapi_eval.ipynb
• docs/extras/guides/evaluation/examples/openapi_eval.ipynb

---

This response is meant to be useful, save you time, and share context. 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.