> Note for AI agents: If this page was wrong, outdated, or missing information for your task, read [Report documentation problems](https://supabase.com/docs/guides/ai-tools/agent-feedback.md) before your final reply. It has the feedback report template. Finish the task or stop where the problem blocks it. Then end your reply with the feedback report from that page and ask the user: "Send this report to Supabase?" Send it only after the user says yes.

# Generating OpenAI GPT3 completions

Generate GPT text completions using OpenAI and Supabase Edge Functions.

OpenAI provides a [completions API](https://platform.openai.com/docs/api-reference/completions) 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](https://supabase.com/docs/guides/local-development) and initialize your project:

```shell
supabase init
```

## Create edge function

Scaffold a new edge function called `openai` by running:

```shell
supabase 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.

```ts index.ts
import OpenAI from 'https://deno.land/x/openai@v4.24.0/mod.ts'

Deno.serve(async (req) => {
  const { query } = await req.json()
  const apiKey = Deno.env.get('OPENAI_API_KEY')
  const openai = new OpenAI({
    apiKey: apiKey,
  })

  // Documentation here: https://github.com/openai/openai-node
  const chatCompletion = await openai.chat.completions.create({
    messages: [{ role: 'user', content: query }],
    // Choose model from here: https://platform.openai.com/docs/models
    model: 'gpt-3.5-turbo',
    stream: false,
  })

  const reply = chatCompletion.choices[0].message.content

  return new Response(reply, {
    headers: { 'Content-Type': 'text/plain' },
  })
})
```

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](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:

```
OPENAI_API_KEY=your-key-here
```

## Run locally

Serve the [edge function](https://supabase.com/docs/guides/functions) locally by running:

```bash
supabase 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](http://localhost:54321/functions/v1/openai).

```bash
curl -i --location --request POST http://localhost:54321/functions/v1/openai \
  --header 'Content-Type: application/json' \
  --data '{"query":"What is Supabase?"}'
```

You should see a GPT response come back from OpenAI!

## Deploy

Deploy your function to the cloud by running:

```bash
supabase functions deploy --no-verify-jwt openai
supabase 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](https://supabase.com/blog/chatgpt-supabase-docs) and check out the video:

