OpenAI is developing a new kind of API interaction – OpenAI Assistants – that goes beyond the Completion APIs that we already support through the GPT Task Action.
Why have they provided these new API interaction features?
Companies have suddenly discovered the power of LLMs AI to provide customer service automation to their end-users, utilizing generative AI and the well-known RAG paradigm. The RAG paradigm is a method that employs NLP and generative AI to effortlessly respond to user inquiries through your own Knowledge Bases—whether it’s your website pages, company PDFs/DOCX, or database contents. At Tiledesk, we natively and easily support this capability with our Knowledge Base for Chatbots.
What assistants offer more is an integrated environment that easily and natively supports RAGs (vector indexing and searching of content), data analysis (through the Code interpreter), and invoking custom functions. Moreover, ChatGPT assistants can maintain the conversation history through the concept of “Threads.”
At the time of writing this article, Assistants are still in beta (v2). You can interact with Assistants only through APIs or the Developers Playground, a type of UI to set up and experiment with Assistants (see picture below)
Set up your Assistants as RAG bots
The first great thing you can do with Assistants is respond to your own data using the power of OpenAI. You can add your own PDFs to the Assistant using the Playground. From now on, whenever you ask something of the Assistant, it will reply accordingly, using your own content.
In this example, I uploaded a PDF manual to the File search section.
From now on, the Assistant will always reply based on the contents of the iPhone PDF manual that I uploaded. I can upload up to 100GB per organization, and files up to 512MB in size.
I asked the Assistant, and it replied as I expected.
Now the next step.
Put your Assistants to work.
Great, you’ve grasped the power of OpenAI Assistants. Now, you want to utilize their capabilities. You need to “publish” them on a channel to allow your users to interact with it. A channel is a chat application, for example, a widget on your website or a WhatsApp number, where your end-users seek support. The idea is that you will use your configured Assistant to provide OpenAI-powered support, fed by your documents, on your favourite channel.
After you have successfully configured your Assistant using the OpenAI Playground application, move back to the Tiledesk Design Studio. Open your chatbot. Drag the ChatGPT Assistant directly into the defaultFallback block of your chatbot.
Remove the preconfigured Reply action from the block
Click on the Assistant action to open the detailed view
As you can see, you need the Assistant ID. Copy it from the Playground:
Paste it into the “Assign GPT Assistant” placeholder
Now, connect a Reply block to the “success” endpoint. In this way, the Assistant output attribute “assistantReply” will be sent to the user anytime they ask for something. Select “assistantReply” from the attributes menu.
Now your flow is complete and ready for interactions.
One more thing. Before you start, you need to set up your OpenAI integration in Settings > Integrations. Copy and paste your OpenAI API key into the OpenAI section.
Okay, we are ready to interact with our Assistant. Simply go back to your chatbot and press the play button on the top right corner of the stage.
As you can see, you’ve started interacting with the Assistant. You provide questions, and it provides answers based on the uploaded documents.
This is the power of customer service automation that truly saves costs and boosts CSAT for your business.
Let’s experiment with our ChatGPT Assistant and share your feedback! Feel free to write to support@tiledesk.com.
