A simple interface for making and querying Pinecone vector databases. Use OpenAI embeddings to vectorize and search

Create keys for Pinecone and OpenAI, and store then in your environment variables. Then,

import PineconeIndex from "https://esm.town/v/peterqliu/PineconeIndex/main.tsx";

// set up your environment variables
const pineconeKey = Deno.env.get("PINECONE_KEY");
const modelToken = Deno.env.get("OPENAI_KEY");

const index = new PineconeIndex({
  name: "2025-all-docs",
  model: "text-embedding-ada-002",
  dimensions: 1536,
  pineconeKey,
  modelToken,
});

// run this only once
await index.create();

// Using the methods directly (see bottom of README for HTTP access)
const results = await index.query("machine learning applications");
await index.upsert(["Document 1", "Document 2"]);
await index.empty();

The options passed to the constructor are a thin abstraction over options for new Pinecone().createIndex(). For more detail, consult that documentation.

Create a new Pinecone index (if it doesn't exist).

Find the most relevant documents for your query.

const results = await index.query("your search text");

Returns:

{
  "matches": [
    {
      "id": "doc-123",
      "score": 0.95,
      "metadata": {
        "text": "Machine learning is transforming..."
      }
    }
  ]
}

Items### index.upsert(items)

Upload new documents to your index. Accepts an array of strings or objects containing metadata.

If passing objects, each object must contain a text property which will be embedded. The entire object is stored as metadata, which is useful for filtering or retrieving structured data later.

// Option 1: Simple array of strings
await index.upsert([
  "First document content",
  "Second document content",
]);

// Option 2: Array of objects (entire object is saved as metadata)
await index.upsert([
  { 
    text: "The quick brown fox", 
    category: "animals", 
    source: "book" 
  },
  { 
    text: "Jumped over the lazy dog", 
    category: "animals", 
    rating: 5 
  }
]);

Remove all documents from your index.

await index.empty();

PineconeIndex also provides handleRequest as a convenience method to access your indices via HTTP. This is useful when accessing the index from other vals, especially without sharing Pinecone and OpenAI credentials.

First instantiate your index:

const index = new PinconeIndex({...});

export default async function (req: Request): Promise<Response> {
  return await index.handleRequest(req);
}

In vals that need to Operations are determined by the URL path (/query, /upsert, /clear). The first segment after the domain specifies the operation to perform.

const API_URL = "https://index-val-url.web.val.run";

// Search for documents: append the desired operation to the url
const searchResponse = await fetch(`${API_URL}/query/`, {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify("artificial intelligence trends"),
});