Author: Shubham Tidke
Chatbots have become an integral part of customer service and sales operations. They are used to automate customer service tasks, provide personalized customer experiences, and even help close sales.
Salesforce is one of the most popular customer relationship management (CRM) platforms, and it is no surprise that many companies are looking to integrate their chatbot with Salesforce. By integrating a chatbot with Salesforce, companies can leverage the power of both platforms to create a more efficient and effective customer service and sales process.
ChatGPT is a powerful chatbot platform that can be easily integrated with Salesforce. ChatGPT provides a comprehensive set of features that allow companies to create powerful chatbots that can be used to automate customer service tasks, provide personalized customer experiences, and even help close sales.
Creating a chatbot using ChatGPT is a great way to provide customers with quick and efficient service. It can help you automate mundane tasks and provide customers with personalized recommendations. With the help of Salesforce, you can easily integrate your chatbot with your existing customer service and sales strategies.
Pre-requisite:
- Salesforce account ( https://developer.salesforce.com/signup )
- ChatGPT account ( https://chat.apps.openai.com/auth/login )
Steps:
1. Add chatbot as a utility item in your salesforce app.
In your salesforce org, setup -> App Manager -> Utility Items -> Add Utility item and search chatbot. Click on it and save. If you cannot find it, skip the step and follow step 2.
2. Get Access Key for chatGPT API
To access the chatGPT endpoint, we need an API key that is unique for every user. API Key can be generated using the link: https://platform.openai.com/account/api-keys
3. Add API Key to Salesforce.
- Add the API Key as a custom label key in your Salesforce org.
- Go to Setup -> Custom Labels -> New Custom Label, and provide any name with API KEY as value.
The idea behind keeping the API key as the custom label is to keep your API key hidden. We can easily import this API key into our lightning web component without exposing it publicly.
4. Create a Lightning Web Component to customize the chatbot.
Using VS Code, create an LWC component. We will be using the completions endpoint which will answer the questions provided to the chatbot as input.
API documentation link: https://platform.openai.com/docs/models/content-filter
Create an input field to take input from users in the chatbot with a button on click of which we can load our results. A spinner is also added in HTML, which will load on the screen till the API returns the result data.
Add the targets in the meta.xml file of your component and deploy the code.
In order to deploy it in the utility bar, we need to add <target>lightning__UtilityBar</target>
When you deploy the code, go to setup -> App Manager -> Utility Items -> Add Utility item and in the search box find your custom component’s name. Add the component and save. It will look something like this:
In JS, import the custom label we created in the org to store the API key. Create a variable to store text input received from the user via the chatbot and in the event type of the button on the chatbot we will send a POST request to the ChatGPT endpoint.
There are a few headers that we need to pass to the endpoint:
1. Prompt: it will be the input text (a question we will be asking the chatbot)
2. Model: OPEN API supports different models which have different capabilities, here we will use “text-davinci-003” which understands natural language and generates longer inputs (4000 tokens).
3. Temperature: Using this header we can customize the creativity of the answer. The lower the temperature, the more straightforward the answers.
4. max_tokens: max limit of pieces of words we will be using. 1000 tokens are around 750 words.
5. Stream: to stream back partial progress. If set, tokens will be sent as data-only server-sent events as they become available, with the stream terminated by a data: [DONE] message.
6. top_p: An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered.
Using fetch we will send a request to the endpoint.
After deploying, add the endpoint as CSP trusted sites in your org.
Once done, you are going to run your chatbot supported by ChatGPT. You can get the full code here. Below are some snippets.
