Integration Migration April 13, 2026

Migrate from Pinecone to REM Labs

Pinecone is a vector database. REM Labs is an AI memory system. This guide walks through exporting your Pinecone vectors and importing them into REM Labs, where they gain full-text search, entity extraction, temporal decay, and neural reranking on top of vector similarity.

Why Migrate

Multi-signal retrieval -- REM fuses vector, full-text, and entity graph results instead of relying on embeddings alone
No embedding management -- REM generates and indexes embeddings automatically; you send raw text
Memory features -- temporal decay, consolidation, namespaces, and Dream Engine are not available in a vector DB
Lower cost -- no per-pod pricing; REM uses a simple request + storage model

Step 1: Export from Pinecone

Use the Pinecone client to fetch all vectors with their metadata:

import { Pinecone } from "@pinecone-database/pinecone";
import fs from "fs";

const pc = new Pinecone({ apiKey: process.env.PINECONE_API_KEY });
const index = pc.index("your-index");

// List all vector IDs
const allIds = [];
for await (const ids of index.listPaginated()) {
  allIds.push(...ids.vectors.map(v => v.id));
}

// Fetch vectors in batches of 100
const exported = [];
for (let i = 0; i < allIds.length; i += 100) {
  const batch = allIds.slice(i, i + 100);
  const response = await index.fetch(batch);
  for (const [id, vec] of Object.entries(response.records)) {
    exported.push({
      id,
      text: vec.metadata?.text || "",
      metadata: vec.metadata || {}
    });
  }
}

fs.writeFileSync("pinecone-export.json", JSON.stringify(exported, null, 2));
console.log(`Exported ${exported.length} vectors`);

Step 2: Import into REM Labs

Read the export and push each record to REM Labs. You send raw text -- REM handles embedding automatically:

import { RemClient } from "@remlabs/sdk";
import fs from "fs";

const rem = new RemClient({ apiKey: process.env.REMLABS_API_KEY });
const data = JSON.parse(fs.readFileSync("pinecone-export.json", "utf-8"));

let imported = 0;
for (const record of data) {
  if (!record.text) continue;

  await rem.remember({
    content: record.text,
    namespace: "pinecone-import",
    metadata: {
      pinecone_id: record.id,
      ...record.metadata
    },
    tags: record.metadata?.tags || []
  });

  imported++;
  if (imported % 100 === 0) {
    console.log(`Imported ${imported}/${data.length}`);
  }
}

console.log(`Migration complete: ${imported} memories imported`);

Step 3: Verify the Migration

Run a few test queries to compare results between Pinecone and REM:

// Test query against REM Labs
const results = await rem.recall({
  query: "What is our refund policy?",
  namespace: "pinecone-import",
  limit: 5
});

results.forEach((m, i) => {
  console.log(`${i + 1}. [${m.score.toFixed(3)}] ${m.content.slice(0, 100)}...`);
});

Step 4: Update Your Application Code

Replace Pinecone query calls with REM Labs recall. The key difference: you send natural language queries instead of pre-computed embeddings.

// Before (Pinecone)
const embedding = await openai.embeddings.create({
  model: "text-embedding-3-small",
  input: userQuery
});
const results = await index.query({
  vector: embedding.data[0].embedding,
  topK: 5,
  includeMetadata: true
});

// After (REM Labs)
const results = await rem.recall({
  query: userQuery,
  namespace: "pinecone-import",
  limit: 5
});

No embedding calls needed: REM Labs generates embeddings internally. This eliminates the OpenAI embeddings API call from your query path, reducing both latency and cost.

Move beyond vector-only search

Free tier. No embedding management. Multi-signal retrieval out of the box.

Get Started