Inserting and Updating Vectors: Upsert Operations

In Pinecone, you do not write traditional SQL INSERT commands. Instead, you perform Upsert operations. If a vector with the provided ID already exists, it is updated; otherwise, a new record is created.

1. Structuring the Vector Payload

Every record inserted into Pinecone contains:

• id: A unique string identifier.
• values: An array of decimal numbers representing the vector embedding.
• metadata: An optional JSON object containing data fields (such as text content or categories).

2. Executing an Upsert Operation

To upload vector records, reference your target index and call the .upsert() method:

import { pc } from "../lib/pinecone";

async function populateIndex() {
  const index = pc.index("knowledge-base");

  console.log("Preparing vector payloads...");

  const records = [
    {
      id: "doc_1",
      values: [0.015, -0.024, 0.089, 0.012], // Reduced dimension representation
      metadata: {
        title: "Introduction to React",
        category: "frontend",
        lines: 45,
      },
    },
    {
      id: "doc_2",
      values: [-0.008, 0.045, -0.012, 0.096],
      metadata: {
        title: "State Management with Zustand",
        category: "frontend",
        lines: 32,
      },
    },
  ];

  await index.upsert(records);
  console.log("Vectors uploaded successfully!");
}

3. Data Type Constraints

• ID size: Maximum of 64 characters.
• Vector values: Must match the dimensions defined when the index was created.
• Metadata size: Maximum of 40KB per vector record.