Skip to main content

Vercel Postgres

LangChain.js supports using the @vercel/postgres package to use generic Postgres databases as vector stores, provided they support the pgvector Postgres extension.

This integration is particularly useful from web environments like Edge functions.

Setup

To work with Vercel Postgres, you need to install the @vercel/postgres package:

npm install @vercel/postgres
npm install @langchain/community

This integration automatically connects using the connection string set under process.env.POSTGRES_URL. You can also pass a connection string manually like this:

const vectorstore = await VercelPostgres.initialize(new OpenAIEmbeddings(), {
postgresConnectionOptions: {
connectionString:
"postgres://<username>:<password>@<hostname>:<port>/<dbname>",
},
});

Connecting to Vercel Postgres

A simple way to get started is to create a serverless Vercel Postgres instance. If you're deploying to a Vercel project with an associated Vercel Postgres instance, the required POSTGRES_URL environment variable will already be populated in hosted environments.

Connecting to other databases

If you prefer to host your own Postgres instance, you can use a similar flow to LangChain's PGVector vectorstore integration and set the connection string either as an environment variable or as shown above.

Usage

import { CohereEmbeddings } from "@langchain/cohere";
import { VercelPostgres } from "@langchain/community/vectorstores/vercel_postgres";

// Config is only required if you want to override default values.
const config = {
// tableName: "testvercelvectorstorelangchain",
// postgresConnectionOptions: {
// connectionString: "postgres://<username>:<password>@<hostname>:<port>/<dbname>",
// },
// columns: {
// idColumnName: "id",
// vectorColumnName: "vector",
// contentColumnName: "content",
// metadataColumnName: "metadata",
// },
};

const vercelPostgresStore = await VercelPostgres.initialize(
new CohereEmbeddings(),
config
);

const docHello = {
pageContent: "hello",
metadata: { topic: "nonsense" },
};
const docHi = { pageContent: "hi", metadata: { topic: "nonsense" } };
const docMitochondria = {
pageContent: "Mitochondria is the powerhouse of the cell",
metadata: { topic: "science" },
};

const ids = await vercelPostgresStore.addDocuments([
docHello,
docHi,
docMitochondria,
]);

const results = await vercelPostgresStore.similaritySearch("hello", 2);
console.log(results);
/*
[
Document { pageContent: 'hello', metadata: { topic: 'nonsense' } },
Document { pageContent: 'hi', metadata: { topic: 'nonsense' } }
]
*/

// Metadata filtering
const results2 = await vercelPostgresStore.similaritySearch(
"Irrelevant query, metadata filtering",
2,
{
topic: "science",
}
);
console.log(results2);
/*
[
Document {
pageContent: 'Mitochondria is the powerhouse of the cell',
metadata: { topic: 'science' }
}
]
*/

// Metadata filtering with IN-filters works as well
const results3 = await vercelPostgresStore.similaritySearch(
"Irrelevant query, metadata filtering",
3,
{
topic: { in: ["science", "nonsense"] },
}
);
console.log(results3);
/*
[
Document {
pageContent: 'hello',
metadata: { topic: 'nonsense' }
},
Document {
pageContent: 'hi',
metadata: { topic: 'nonsense' }
},
Document {
pageContent: 'Mitochondria is the powerhouse of the cell',
metadata: { topic: 'science' }
}
]
*/

// Upserting is supported as well
await vercelPostgresStore.addDocuments(
[
{
pageContent: "ATP is the powerhouse of the cell",
metadata: { topic: "science" },
},
],
{ ids: [ids[2]] }
);

const results4 = await vercelPostgresStore.similaritySearch(
"What is the powerhouse of the cell?",
1
);
console.log(results4);
/*
[
Document {
pageContent: 'ATP is the powerhouse of the cell',
metadata: { topic: 'science' }
}
]
*/

await vercelPostgresStore.delete({ ids: [ids[2]] });

const results5 = await vercelPostgresStore.similaritySearch(
"No more metadata",
2,
{
topic: "science",
}
);
console.log(results5);
/*
[]
*/

// Remember to call .end() to close the connection!
await vercelPostgresStore.end();

API Reference:


Was this page helpful?


You can also leave detailed feedback on GitHub.