AI & Vectors

Vector search with Next.js and OpenAI

Learn how to build a ChatGPT-style doc search powered by Next.js, OpenAI, and Supabase.

While our Headless Vector search provides a toolkit for generative Q&A, in this tutorial we'll go more in-depth, build a custom ChatGPT-like search experience from the ground-up using Next.js. You will:

  1. Convert your markdown into embeddings using OpenAI.
  2. Store you embeddings in Postgres using pgvector.
  3. Deploy a function for answering your users' questions.

You can read our Supabase Clippy blog post for a full example.

We assume that you have a Next.js project with a collection of .mdx files nested inside your pages directory. We will start developing locally with the Supabase CLI and then push our local database changes to our hosted Supabase project. You can find the full Next.js example on GitHub.

Create a project

  1. Create a new project in the Supabase Dashboard.
  2. Enter your project details.
  3. Wait for the new database to launch.

Prepare the database

Let's prepare the database schema. We can use the "OpenAI Vector Search" quickstart in the SQL Editor, or you can copy/paste the SQL below and run it yourself.

  1. Go to the SQL Editor page in the Dashboard.
  2. Click OpenAI Vector Search.
  3. Click Run.

Pre-process the knowledge base at build time

With our database set up, we need to process and store all .mdx files in the pages directory. You can find the full script here, or follow the steps below:

1

Generate Embeddings

Create a new file lib/generate-embeddings.ts and copy the code over from GitHub.

1
curl \
2
https://raw.githubusercontent.com/supabase-community/nextjs-openai-doc-search/main/lib/generate-embeddings.ts \
3
-o "lib/generate-embeddings.ts"
2

Set up environment variables

We need some environment variables to run the script. Add them to your .env file and make sure your .env file is not committed to source control! You can get your local Supabase credentials by running supabase status.

1
NEXT_PUBLIC_SUPABASE_URL=
2
NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY=
3
SUPABASE_SERVICE_ROLE_KEY=
4
5
# Get your key at https://platform.openai.com/account/api-keys
6
OPENAI_API_KEY=
3

Run script at build time

Include the script in your package.json script commands to enable Vercel to automatically run it at build time.

1
"scripts": {
2
  "dev": "next dev",
3
  "build": "pnpm run embeddings && next build",
4
  "start": "next start",
5
  "embeddings": "tsx lib/generate-embeddings.ts"
6
},

Create text completion with OpenAI API

Anytime a user asks a question, we need to create an embedding for their question, perform a similarity search, and then send a text completion request to the OpenAI API with the query and then context content merged together into a prompt.

All of this is glued together in a Vercel Edge Function, the code for which can be found on GitHub.

1

Create Embedding for Question

In order to perform similarity search we need to turn the question into an embedding.

1
const embeddingResponse = await fetch('https://api.openai.com/v1/embeddings', {
2
  method: 'POST',
3
  headers: {
4
    Authorization: `Bearer ${openAiKey}`,
5
    'Content-Type': 'application/json',
6
  },
7
  body: JSON.stringify({
8
    model: 'text-embedding-ada-002',
9
    input: sanitizedQuery.replaceAll('\n', ' '),
10
  }),
11
})
12
13
if (embeddingResponse.status !== 200) {
14
  throw new ApplicationError('Failed to create embedding for question', embeddingResponse)
15
}
16
17
const {
18
  data: [{ embedding }],
19
} = await embeddingResponse.json()
2

Perform similarity search

Using the embeddingResponse we can now perform similarity search by performing an remote procedure call (RPC) to the database function we created earlier.

1
const { error: matchError, data: pageSections } = await supabaseClient.rpc(
2
  'match_page_sections',
3
  {
4
    embedding,
5
    match_threshold: 0.78,
6
    match_count: 10,
7
    min_content_length: 50,
8
  }
9
)
3

Perform text completion request

With the relevant content for the user's question identified, we can now build the prompt and make a text completion request via the OpenAI API.

If successful, the OpenAI API will respond with a text/event-stream response that we can forward to the client where we'll process the event stream to smoothly print the answer to the user.

1
const prompt = codeBlock`
2
  ${oneLine`
3
    You are a very enthusiastic Supabase representative who loves
4
    to help people! Given the following sections from the Supabase
5
    documentation, answer the question using only that information,
6
    outputted in markdown format. If you are unsure and the answer
7
    is not explicitly written in the documentation, say
8
    "Sorry, I don't know how to help with that."
9
  `}
10
11
  Context sections:
12
  ${contextText}
13
14
  Question: """
15
  ${sanitizedQuery}
16
  """
17
18
  Answer as markdown (including related code snippets if available):
19
`
20
21
const completionOptions: CreateCompletionRequest = {
22
  model: 'gpt-3.5-turbo-instruct',
23
  prompt,
24
  max_tokens: 512,
25
  temperature: 0,
26
  stream: true,
27
}
28
29
const response = await fetch('https://api.openai.com/v1/completions', {
30
  method: 'POST',
31
  headers: {
32
    Authorization: `Bearer ${openAiKey}`,
33
    'Content-Type': 'application/json',
34
  },
35
  body: JSON.stringify(completionOptions),
36
})
37
38
if (!response.ok) {
39
  const error = await response.json()
40
  throw new ApplicationError('Failed to generate completion', error)
41
}
42
43
// Proxy the streamed SSE response from OpenAI
44
return new Response(response.body, {
45
  headers: {
46
    'Content-Type': 'text/event-stream',
47
  },
48
})

Display the answer on the frontend

In a last step, we need to process the event stream from the OpenAI API and print the answer to the user. The full code for this can be found on GitHub.

1
const handleConfirm = React.useCallback(
2
  async (query: string) => {
3
    setAnswer(undefined)
4
    setQuestion(query)
5
    setSearch('')
6
    dispatchPromptData({ index: promptIndex, answer: undefined, query })
7
    setHasError(false)
8
    setIsLoading(true)
9
10
    const eventSource = new SSE(`api/vector-search`, {
11
      headers: {
12
        apikey: process.env.NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY ?? '',
13
        Authorization: `Bearer ${process.env.NEXT_PUBLIC_SUPABASE_PUBLISHABLE_KEY}`,
14
        'Content-Type': 'application/json',
15
      },
16
      payload: JSON.stringify({ query }),
17
    })
18
19
    function handleError<T>(err: T) {
20
      setIsLoading(false)
21
      setHasError(true)
22
      console.error(err)
23
    }
24
25
    eventSource.addEventListener('error', handleError)
26
    eventSource.addEventListener('message', (e: any) => {
27
      try {
28
        setIsLoading(false)
29
30
        if (e.data === '[DONE]') {
31
          setPromptIndex((x) => {
32
            return x + 1
33
          })
34
          return
35
        }
36
37
        const completionResponse: CreateCompletionResponse = JSON.parse(e.data)
38
        const text = completionResponse.choices[0].text
39
40
        setAnswer((answer) => {
41
          const currentAnswer = answer ?? ''
42
43
          dispatchPromptData({
44
            index: promptIndex,
45
            answer: currentAnswer + text,
46
          })
47
48
          return (answer ?? '') + text
49
        })
50
      } catch (err) {
51
        handleError(err)
52
      }
53
    })
54
55
    eventSource.stream()
56
57
    eventSourceRef.current = eventSource
58
59
    setIsLoading(true)
60
  },
61
  [promptIndex, promptData]
62
)

Learn more

Want to learn more about the awesome tech that is powering this?