This is code for building a GPT-3 powered knowledge base application using Next.js. It's adapted from the example in the book Exploring GPT-3. To deploy and run this code you'll need the following:
- Access to the OpenAI API - Sign up for an account
- An account on vercel.com (a free account will work fine)
- Watch the tutorial video.
- Create and/or copy an API key from the OpenAI API Keys Page.
- Click the 'Deploy with Vercel' button below to begin deploying this app to your vercel.com account.
- Enter your OpenAI API key for the environment variable named
OPENAI_API_KEY. - Add your own content to the documents defined in the
routes/answers.jsonfile. - Test the app by visiting the Vercel generated app URL.
By default the code uses a JavaScript array of documents in routes/answers.json for the knowledge base. Alternately, you can store documents in an answers file. There is an example answers file named answers.jsonl in the root, along with a file named upload.js that provides code you can use to upload an answers file to the OpenAI API. If you use the answers file, you'll also need to setup a secret / environment variable named ANSWERS_FILE with the OpenAI filename for the value.
An array of documents is used by default because because it's simpler. However, there are two primary advantages to using a documents file. First, you can include more than 200 documents. Second, the the API will throw a 404 error if there isn't a matching result. Meaning, the app could be setup to log questions that should be added to the documents - unanswered questions. This is not currently a feature but is planned for a future release.
For more details on using answer files, see the OpenAI Documentation.
If you have questions or comments, please post them on the Exploring GPT-3 Discord Server or in the GitHub Issues page for this repository. Thanks so much!