supabase/apps/docs/content/guides/functions/examples/semantic-search.mdx

---
id: 'function-ai-models'
title: 'Semantic Search'
description: 'Semantic Search with pgvector and Supabase Edge Functions'
subtitle: 'Semantic Search with pgvector and Supabase Edge Functions'
tocVideo: 'w4Rr_1whU-U'
---

[Semantic search](/docs/guides/ai/semantic-search) interprets the meaning behind user queries rather than exact [keywords](/docs/guides/ai/keyword-search). It uses machine learning to capture the intent and context behind the query, handling language nuances like synonyms, phrasing variations, and word relationships.

Since Supabase Edge Runtime [v1.36.0](https://github.com/supabase/edge-runtime/releases/tag/v1.36.0) you can run the [`gte-small` model](https://huggingface.co/Supabase/gte-small) natively within Supabase Edge Functions without any external dependencies! This allows you to generate text embeddings without calling any external APIs!

In this tutorial you're implementing three parts:

1. A [`generate-embedding`](https://github.com/supabase/supabase/tree/master/examples/ai/edge-functions/supabase/functions/generate-embedding/index.ts) database webhook edge function which generates embeddings when a content row is added (or updated) in the [`public.embeddings`](https://github.com/supabase/supabase/tree/master/examples/ai/edge-functions/supabase/migrations/20240408072601_embeddings.sql) table.
2. A [`query_embeddings` Postgres function](https://github.com/supabase/supabase/tree/master/examples/ai/edge-functions/supabase/migrations/20240410031515_vector-search.sql) which allows us to perform similarity search from an Edge Function via [Remote Procedure Call (RPC)](/docs/guides/database/functions?language=js).
3. A [`search` edge function](https://github.com/supabase/supabase/tree/master/examples/ai/edge-functions/supabase/functions/search/index.ts) which generates the embedding for the search term, performs the similarity search via RPC function call, and returns the result.

You can find the complete example code on [GitHub](https://github.com/supabase/supabase/tree/master/examples/ai/edge-functions)

### Create the database table and webhook

Given the [following table definition](https://github.com/supabase/supabase/blob/master/examples/ai/edge-functions/supabase/migrations/20240408072601_embeddings.sql):

```sql
create extension if not exists vector with schema extensions;

create table embeddings (
  id bigint primary key generated always as identity,
  content text not null,
  embedding extensions.vector (384)
);
alter table embeddings enable row level security;

create index on embeddings using hnsw (embedding vector_ip_ops);
```

You can deploy the [following edge function](https://github.com/supabase/supabase/blob/master/examples/ai/edge-functions/supabase/functions/generate-embedding/index.ts) as a [database webhook](/docs/guides/database/webhooks) to generate the embeddings for any text content inserted into the table:

```ts
import { withSupabase } from 'npm:@supabase/server@^1'

const model = new Supabase.ai.Session('gte-small')

// Triggered by a Database Webhook, which authenticates with a secret key.
// Deploy with `verify_jwt = false`.
export default {
  fetch: withSupabase({ auth: 'secret' }, async (req, ctx) => {
    const payload: WebhookPayload = await req.json()
    const { content, id } = payload.record

    // Generate embedding.
    const embedding = await model.run(content, {
      mean_pool: true,
      normalize: true,
    })

    // Store in database.
    const { error } = await ctx.supabaseAdmin
      .from('embeddings')
      .update({ embedding: JSON.stringify(embedding) })
      .eq('id', id)
    if (error) console.warn(error.message)

    return Response.json({ ok: true })
  }),
}
```

## Create a Database Function and RPC

With the embeddings now stored in your Postgres database table, you can query them from Supabase Edge Functions by using [Remote Procedure Calls (RPC)](/docs/guides/database/functions?language=js).

Given the [following Postgres Function](https://github.com/supabase/supabase/blob/master/examples/ai/edge-functions/supabase/migrations/20240410031515_vector-search.sql):

```sql
-- Matches document sections using vector similarity search on embeddings
--
-- Returns a setof embeddings so that we can use PostgREST resource embeddings (joins with other tables)
-- Additional filtering like limits can be chained to this function call
create or replace function query_embeddings(embedding extensions.vector(384), match_threshold float)
returns setof embeddings
language plpgsql
as $$
begin
  return query
  select *
  from embeddings

  -- The inner product is negative, so we negate match_threshold
  where embeddings.embedding <#> embedding < -match_threshold

  -- Our embeddings are normalized to length 1, so cosine similarity
  -- and inner product will produce the same query results.
  -- Using inner product which can be computed faster.
  --
  -- For the different distance functions, see https://github.com/pgvector/pgvector
  order by embeddings.embedding <#> embedding;
end;
$$;
```

## Query vectors in Supabase Edge Functions

You can use `supabase-js` to first generate the embedding for the search term and then invoke the Postgres function to find the relevant results from your stored embeddings, right from your [Supabase Edge Function](https://github.com/supabase/supabase/blob/master/examples/ai/edge-functions/supabase/functions/search/index.ts):

```ts
import { withSupabase } from 'npm:@supabase/server@^1'

const model = new Supabase.ai.Session('gte-small')

export default {
  fetch: withSupabase({ auth: 'user' }, async (req, ctx) => {
    const { search } = await req.json()
    if (!search) return Response.json({ error: 'Please provide a search param!' }, { status: 400 })
    // Generate embedding for search term.
    const embedding = await model.run(search, {
      mean_pool: true,
      normalize: true,
    })

    // Query embeddings.
    const { data: result, error } = await ctx.supabase
      .rpc('query_embeddings', {
        embedding,
        match_threshold: 0.8,
      })
      .select('content')
      .limit(3)
    if (error) {
      return Response.json({ error: error.message }, { status: 500 })
    }

    return Response.json({ search, result })
  }),
}
```

You now have AI powered semantic search set up without any external dependencies! Just you, pgvector, and Supabase Edge Functions!