Generating OpenAI GPT3 completions
Generate GPT text completions using OpenAI and Supabase Edge Functions.
OpenAI provides a completions API that allows you to use their generative GPT models in your own applications.
OpenAI's API is intended to be used from the server-side. Supabase offers Edge Functions to make it easy to interact with third party APIs like OpenAI.
Setup Supabase project
If you haven't already, install the Supabase CLI and initialize your project:
_10supabase init
Create edge function
Scaffold a new edge function called openai
by running:
_10supabase functions new openai
A new edge function will now exist under ./supabase/functions/openai/index.ts
.
We'll design the function to take your user's query (via POST request) and forward it to OpenAI's API.
Note that we are setting stream
to false
which will wait until the entire response is complete before returning. If you wish to stream GPT's response word-by-word back to your client, set stream
to true
.
Create OpenAI key
You may have noticed we were passing OPENAI_API_KEY
in the Authorization header to OpenAI. To generate this key, go to https://platform.openai.com/account/api-keys and create a new secret key.
After getting the key, copy it into a new file called .env.local
in your ./supabase
folder:
_10OPENAI_API_KEY=your-key-here
Run locally
Serve the edge function locally by running:
_10supabase functions serve --env-file ./supabase/.env.local --no-verify-jwt
Notice how we are passing in the .env.local
file.
Use cURL or Postman to make a POST request to http://localhost:54321/functions/v1/openai.
_10curl -i --location --request POST http://localhost:54321/functions/v1/openai \_10 --header 'Content-Type: application/json' \_10 --data '{"query":"What is Supabase?"}'
You should see a GPT response come back from OpenAI!
Deploy
Deploy your function to the cloud by runnning:
_10supabase functions deploy --no-verify-jwt openai_10supabase secrets set --env-file ./supabase/.env.local
Go deeper
If you're interesting in learning how to use this to build your own ChatGPT, read the blog post and check out the video: