Technical Articles
Integration of SAP CPI(BTP IS) with ChatGPT
Integration of SAP CPI(BTP IS) with ChatGPT
Introduction
What is ChatGPT?
ChatGPT is a language model developed by OpenAI. It is designed to understand natural language, generate human-like responses to a wide range of questions, and carry out various language-related tasks. It has been trained on a large corpus of text data and can understand a broad range of topics. Its purpose is to assist users in their quest for knowledge and provide them with informative and useful responses.
What is CPI(BTP-IS)?
CPI(BTP-IS) is a set of services and tools provided by SAP on its cloud-based Business Technology Platform (BTP) to enable integration between different systems, applications, and data sources. The key benefit of CPI(BTP IS) is that it enables organizations to quickly and easily integrate their systems, data, and applications without the need for extensive coding or custom development. This helps to streamline business processes, reduce costs, and improve operational efficiency.
How ChatGPT can be integrated?
ChatGPT can be integrated using APIs (Application Programming Interfaces) provided by OpenAI. The API can be used to send text inputs to ChatGPT and receive responses in real-time. Chatbot Platform Integration and Custom Integration using libraries and SDK also possible.
How CPI integrates with ChatGPT?
CPI interacts ChatGPT through HTTP requests using API Keys for authentication. ChatGPT has multiple APIs depending on the usage or self learning basis like Models, Completions, Edits, Images, Create Image Variations, Embeddings, Files, Fine-tunes and Moderations APIs
Note: All the available ChatGPT APIs can be integrated with SAP CPI but for demonstration will use the Completion API.
Integration of SAP CPI(BTP IS) with ChatGPT
Prerequisites:
- SAP BTP IS Tenant Access
- ChatGPT openai platform Access
- Postman Tool for testing
Step-1: Create a new secret key on ChatGPT
After login to the account on https://platform.openai.com/account/api-keys goto API Keys and click on Create new secret key.
Image source from my account on platform.openai.com
Step-2: Download the security certificate from respective ChatGPT API
For demo we will use the API “ https://api.openai.com/v1/completions” from other available APIs of ChatGPT.
Open the API url in any browser and click on the lock(HTTPS) icon from the address bar and select show certificate.
Export the root certificate and save to your local desktop.
Image source from my account on platform.openai.com
Step-3: Upload the root security certificate in the SAP BTP IS(CPI)=>KeyStore
Login to BTP account and redirect to your Integration Suite home page. From the left side pane select the monitor artifacts options and open the Keystore page.
Click on Add dropdown option from right side top pane and select the Certificate option and upload the already downloaded security certificate from Step-1.
Image source from my trail account on SAP BTP IS
Step-4: Create a new Integration Flow under the desired package with appropriate naming convention.
From the left side pane design artifacts option select or create a package and create a new IFLOW.
We need to create a scenario of HTTPS sender adapter to HTTP receiver adapter with intermediate call on chatGPT API .
Step-5: Configure a sender HTTPS adapter as we will test it from Postman tool.
As per the requirement we can have any sender adapter which can able to provide the expected input text.
Image source from my trail account on SAP BTP IS
Provide the desired Address path which will be added to the deployed tenant URL.
Step-6: Configure the JSON to XML converter as the input text from Postman will be provided as JSON.
As per the requirement we can have any format input can be sent from sender system which can be modified accordingly.
Image source from my trail account on SAP BTP IS
Remove the Namespace Mapping if selected.
In our case JSON to XML only is used as input from POSTMAN will be JSON and for further processing of exchange properties we will use XML .
Step-7: Configure the Content Modifier to set the Message Header, Exchange Property and Message Body.
The ChatGPT HTTP API will be called with Authorization and ContentType values in the Message Header
So accordingly provide the Header constant in the Content Modifier with Content-Type as application/json
And Authorization will be the Bearer (space) API key generated in the step-1
Image source from my trail account on SAP BTP IS
The sample input from source will be in the below JSON format
{
“prompt”: “Input Text here”
}
In order to capture the Input “Prompt” value and prepare complete request to API the exchange property will be configured with a “input” referring to the Xpath of the JSON converted XML field value.
Image source from my trail account on SAP BTP IS
The original API HTTP call will be expecting the actual source input with its respective parameters. Those parameters and source input will be formulated in the Message Body.
Image source from my trail account on SAP BTP IS
Each ChatGPT API call will be processed based on the model name provided. In our case will provide the latest model “text-davinci-003” which is the most capable GPT-3 model. Can do any task the other models can do, often with higher quality, longer output and better instruction-following. And supports inserting completions within text. Also its trained with the internet content available upto Jun2021.
Step-8: Configure a Request-reply with HTTP adapter receiver connection to ChatGPT API.
Provide the mandatory field values like Address of the ChatGPT API, Method as POST, Authentication as Client Certificate and the Request/Response headers. The Request headers can be given as * or Authorization|Content-Type etc.
Image source from my trail account on SAP BTP IS
Step-9: Handling the API JSON response.
When any ChatGPT API is invoked with JSON request then there will be a JSON response which can be converted to XML to produce the desired field output.
Configure the JSON to XML converter to select the respective XML field values to be populated in the output.
Image source from my trail account on SAP BTP IS
Step-10: Configure the Content Modifier for converted XML payload.
Create a “output” property referring to the XML text field which will be only populated as final output.
Image source from my trail account on SAP BTP IS
Modify the Message body also with the selected property value.
Image source from my trail account on SAP BTP IS
Final Integration Flow will look like:
Image source from my trail account on SAP BTP IS
Once the IFLOW is saved and deployed , go to the Managed Integration Content and get the endpoint which will be used in the source system to trigger a message to CPI.
Image source from my trail account on SAP BTP IS
In order to view the traces in the CPI monitor message processing. Select the log level as Trace instead of default option Info.
Image source from my trail account on SAP BTP IS
Testing in POSTMAN:
Image source from local installed Postman
Provide the CPI tenant endpoint URL for the created IFLOW along with the BTP account credentials in the Postman Authorization section.
For the sample input provided we have received the response accordingly from the chatGPT API.
Monitoring the IFLOW from CPI tenant with the trace enabled.
Image source from my trail account on SAP BTP IS
Image source from my trail account on SAP BTP IS
Conclusion:
Any input text given to the ChatGPT API there will be cutting-edge language models working to understand and generate the response text. With the available ChatGPT APIs we can integrate them with CPI/any middleware to consume and provide more reliable response on top of AI.
If ChatGPT is connected at one side of the SAP CPI(BTP IS) the other side we can have any SAP(ECC, S4/HANA, SuccessFactor,etc.,) or non-SAP systems to achieve the end to end integration.
Disclaimer:
This article is only for informational purpose that CPI can be integrated with the ChatGPT APIs also.
Only a integration of one API has been shown as an example but no limitations in any perspective.
Thank you for sharing the blog. It was a fascinating and inventive read. I'm curious, have you ever been able to successfully implement the solution? I ask because I keep encountering the same error message, "The MPL ID for the failed message is: AGP6qNuZSFQxp9a98qRrC5Vpro5y," and wonder if you've had similar experiences.
Hi Mohan,
Can you please check whether the API Key working and the certificate correctly imported into the keystore. Also make sure that Authorization and Content-Type values should be passed in the header while calling the chatGPT API.
Thanks,
Farooq.
Dear Farooq
Very nice blog neatly presented with a simple example to learn this new topic. Appreciate your effort in writing this blog.
Thanks,
Nagendra
Thank you very much Nagendra.
Regards,
Farooq.
Thanks for the blog! I had some queries around the postman settings for authorization. Is this meant to be 'basic auth' and should it be the 'S' id that is used e.g. to login to BTP/IS.
If you can share your email, I can show you exactly what I mean too please.
Thanks!
JP
From Postman you can have the basic auth providing the same credentials of BTP cockpit login. If the user doesnt have enough roles then create a role collection with MessagingSend role and assign to your user. Also you can try with a service key using clientid and clientsecret.
Regards,
Farooq.
Dear Farooq,
Thanks for the detailed blog. I try to register to chatGBT to get the API key and API, unfortunately, I am getting a message it is not available in your country(KSA).
OpenAI's services are not available in your country.
Regards,
Ayub Ahmed.
If ChatGPT is not available in your country, you can try the methods below to solve the problem. There should be a suitable method for you.
Method 1: Use VPN
Method 2: Use another Web Browser
Method 3: Use ChatGPT in the Private Mode of Your Web Browser
Method 4: Contact ChatGPT Support
Regards,
Farooq
Hi Farroq
First of all I would like to appreciate your efforts for this blog and in detail explanation as well, I tried to replicate the exact composition in my CPI test tenant but getting the below error after performing several test cases. I have mentioned the Header data and maintained Adapter level header properties as *.
com.sap.it.rt.adapter.http.api.exception.HttpResponseException: An internal server error occured: HTTP operation failed invoking https://api.openai.com/v1/completions with statusCode: 400. The MPL ID for the failed message is : AGP_a9AdRrlJGZ2fRoT225R4IGBW
{"model": “text-davinci-003”, "prompt": "who is the PM of INDIA"}
But the issue has been fixed and I am getting proper response with little change. May be this might be useful to others, when we are passing the request it should be arranged in proper JSON like below then getting successful response.
{
"model": "text-davinci-003",
"prompt": "who is the PM of India"
}
Regards
Ch Sudheer
Thanks! I was also getting an error, so it could be something is 'wrong' with this body:
{
"model": “text-davinci-003”,
"prompt": "${property.input}"
}
Note, when I ran the above (without the variable), then it was successful:
{
"model": "text-davinci-003",
"prompt": "What is the capital of the moon?"
}
Farooq - could you please advise what I 'did wrong'. Any updates would be awesome, this is the last step to get this working
Thanks!
Juan
Hello Hadjiandreou,
Check the quotes -
“text-davinci-003” (this is wrong)
"text-davinci-003" (this is correct)
Thanks,
Bhaskar
Thanks for the inputs! I don't see the difference in the quotes - can you share you email please and then I can contact you on this please.
Thanks!
Jose
Using "straight quotes" will be considered as a string by the request body (JSON) if you're using it via postman or any testing tool and throws error if you use “inverted quotes”.
Thanks,
Bhaskar
Thanks so much!
Dear Farooq,
Appreciate your efforts in making an detailed blog. Here is my concern.
while trying to send request using postman, I am not getting the full answer.
Ex: If I am asking "add two numbers using java" , it is not giving full answer. not even one full liner. Please let us know how to sort it out.
I tried using model number *gpt-4* as well, but still no luck.
Thanks in advance.
Thank you so much for the tutorial.
I use this guide with the new GTP-3.5 turbo with the "chat completions" api. If someone is trying to do the same you need to make some modifications.
These are the parts that I modified.
I change the json body according to the official documentation for the chat completions api
update the address for the chat completions api url
the xml is different for the completions api response
so you need to change the source value in the content modifier
That's it, hope this help.
Thank you!
Hi Farooq,
Thank you so much for the detailed post.
I am totally new to CPI , I followe all the steps that mentined however after deploying it , I dont see the end points. Can you please help me to traoubleshoot the issue.
Thank you so much !!