Prepare for the PgBouncer and IPv4 deprecations on 26th January 2024

AI & Vectors

LangChain

LangChain is a popular framework for working with AI, Vectors, and embeddings. LangChain supports using Supabase as a vector store, using the pgvector extension.

Initializing your database

Prepare you database with the relevant tables:

  1. Go to the SQL Editor page in the Dashboard.
  2. Click LangChain in the Quick start section.
  3. Click Run.

Usage

You can now search your documents using any Node.js application. This is intended to be run on a secure server route.


_28
import { SupabaseVectorStore } from 'langchain/vectorstores/supabase'
_28
import { OpenAIEmbeddings } from 'langchain/embeddings/openai'
_28
import { createClient } from '@supabase/supabase-js'
_28
_28
const supabaseKey = process.env.SUPABASE_SERVICE_ROLE_KEY
_28
if (!supabaseKey) throw new Error(`Expected SUPABASE_SERVICE_ROLE_KEY`)
_28
_28
const url = process.env.SUPABASE_URL
_28
if (!url) throw new Error(`Expected env var SUPABASE_URL`)
_28
_28
export const run = async () => {
_28
const client = createClient(url, supabaseKey)
_28
_28
const vectorStore = await SupabaseVectorStore.fromTexts(
_28
['Hello world', 'Bye bye', "What's this?"],
_28
[{ id: 2 }, { id: 1 }, { id: 3 }],
_28
new OpenAIEmbeddings(),
_28
{
_28
client,
_28
tableName: 'documents',
_28
queryName: 'match_documents',
_28
}
_28
)
_28
_28
const resultOne = await vectorStore.similaritySearch('Hello world', 1)
_28
_28
console.log(resultOne)
_28
}

Simple metadata filtering

Given the above match_documents Postgres function, you can also pass a filter parameter to only return documents with a specific metadata field value. This filter parameter is a JSON object, and the match_documents function will use the Postgres JSONB Containment operator @> to filter documents by the metadata field values you specify. See details on the Postgres JSONB Containment operator for more information.


_32
import { SupabaseVectorStore } from 'langchain/vectorstores/supabase'
_32
import { OpenAIEmbeddings } from 'langchain/embeddings/openai'
_32
import { createClient } from '@supabase/supabase-js'
_32
_32
// First, follow set-up instructions above
_32
_32
const privateKey = process.env.SUPABASE_SERVICE_ROLE_KEY
_32
if (!privateKey) throw new Error(`Expected env var SUPABASE_SERVICE_ROLE_KEY`)
_32
_32
const url = process.env.SUPABASE_URL
_32
if (!url) throw new Error(`Expected env var SUPABASE_URL`)
_32
_32
export const run = async () => {
_32
const client = createClient(url, privateKey)
_32
_32
const vectorStore = await SupabaseVectorStore.fromTexts(
_32
['Hello world', 'Hello world', 'Hello world'],
_32
[{ user_id: 2 }, { user_id: 1 }, { user_id: 3 }],
_32
new OpenAIEmbeddings(),
_32
{
_32
client,
_32
tableName: 'documents',
_32
queryName: 'match_documents',
_32
}
_32
)
_32
_32
const result = await vectorStore.similaritySearch('Hello world', 1, {
_32
user_id: 3,
_32
})
_32
_32
console.log(result)
_32
}

Advanced metadata filtering

You can also use query builder-style filtering (similar to how the Supabase JavaScript library works) instead of passing an object. Note that since the filter properties will be in the metadata column, you need to use arrow operators (-> for integer or ->> for text) as defined in Postgrest API documentation and specify the data type of the property (e.g. the column should look something like metadata->some_int_value::int).


_62
import { SupabaseFilterRPCCall, SupabaseVectorStore } from 'langchain/vectorstores/supabase'
_62
import { OpenAIEmbeddings } from 'langchain/embeddings/openai'
_62
import { createClient } from '@supabase/supabase-js'
_62
_62
// First, follow set-up instructions above
_62
_62
const privateKey = process.env.SUPABASE_SERVICE_ROLE_KEY
_62
if (!privateKey) throw new Error(`Expected env var SUPABASE_SERVICE_ROLE_KEY`)
_62
_62
const url = process.env.SUPABASE_URL
_62
if (!url) throw new Error(`Expected env var SUPABASE_URL`)
_62
_62
export const run = async () => {
_62
const client = createClient(url, privateKey)
_62
_62
const embeddings = new OpenAIEmbeddings()
_62
_62
const store = new SupabaseVectorStore(embeddings, {
_62
client,
_62
tableName: 'documents',
_62
})
_62
_62
const docs = [
_62
{
_62
pageContent:
_62
'This is a long text, but it actually means something because vector database does not understand Lorem Ipsum. So I would need to expand upon the notion of quantum fluff, a theoretical concept where subatomic particles coalesce to form transient multidimensional spaces. Yet, this abstraction holds no real-world application or comprehensible meaning, reflecting a cosmic puzzle.',
_62
metadata: { b: 1, c: 10, stuff: 'right' },
_62
},
_62
{
_62
pageContent:
_62
'This is a long text, but it actually means something because vector database does not understand Lorem Ipsum. So I would need to proceed by discussing the echo of virtual tweets in the binary corridors of the digital universe. Each tweet, like a pixelated canary, hums in an unseen frequency, a fascinatingly perplexing phenomenon that, while conjuring vivid imagery, lacks any concrete implication or real-world relevance, portraying a paradox of multidimensional spaces in the age of cyber folklore.',
_62
metadata: { b: 2, c: 9, stuff: 'right' },
_62
},
_62
{ pageContent: 'hello', metadata: { b: 1, c: 9, stuff: 'right' } },
_62
{ pageContent: 'hello', metadata: { b: 1, c: 9, stuff: 'wrong' } },
_62
{ pageContent: 'hi', metadata: { b: 2, c: 8, stuff: 'right' } },
_62
{ pageContent: 'bye', metadata: { b: 3, c: 7, stuff: 'right' } },
_62
{ pageContent: "what's this", metadata: { b: 4, c: 6, stuff: 'right' } },
_62
]
_62
_62
await store.addDocuments(docs)
_62
_62
const funcFilterA: SupabaseFilterRPCCall = (rpc) =>
_62
rpc
_62
.filter('metadata->b::int', 'lt', 3)
_62
.filter('metadata->c::int', 'gt', 7)
_62
.textSearch('content', `'multidimensional' & 'spaces'`, {
_62
config: 'english',
_62
})
_62
_62
const resultA = await store.similaritySearch('quantum', 4, funcFilterA)
_62
_62
const funcFilterB: SupabaseFilterRPCCall = (rpc) =>
_62
rpc
_62
.filter('metadata->b::int', 'lt', 3)
_62
.filter('metadata->c::int', 'gt', 7)
_62
.filter('metadata->>stuff', 'eq', 'right')
_62
_62
const resultB = await store.similaritySearch('hello', 2, funcFilterB)
_62
_62
console.log(resultA, resultB)
_62
}

LangChain supports the concept of a hybrid search, which combines Similarity Search with Full Text Search. Read the official docs to get started: Supabase Hybrid Search.

You can install the LangChain Hybrid Search function though our database.dev package manager.

Resources