Pinecone — Animate | AI Wisdom

← Back to topic
🎞 ANIMATE · Mode 3 of 5 · +100 XP— Pinecone Vector Database
📋Setup
⚙️Ingest
🔄Embed
✅Query
🎯 Next: Test →
Step 1 / 4
Speed
Create Index
pc.index() connects to a Pinecone index — purpose-built vector store💡 Analogy
Like a specialized search engine — you don't build the search infrastructure, just use the API.
Pinecone is a managed vector database. Create an index with dimensions matching your embedding model (1536 for text-embedding-3-small). Connect to it with pc.index().
Console Output
> Pinecone client: authenticated
> Index "knowledge-base": 1536 dims, cosine metric
> Ready to upsert/query
●pinecone.ts
Create IndexL1–7
1import { Pinecone } from '@pinecone-database/pinecone';
2import OpenAI from 'openai';
3 
4const pc = new Pinecone({ apiKey: process.env.PINECONE_API_KEY! });
5const openai = new OpenAI();
6 
7// Create or reuse an index
8const index = pc.index('knowledge-base');
9 
10async function ingest(docs: { id: string; text: string; source: string }[]) {
11  // Generate embeddings in batches
12  const embeddings = await openai.embeddings.create({
13    model: 'text-embedding-3-small',
14    input: docs.map(d => d.text),
15  });
16 
17  // Upsert vectors with metadata
18  await index.upsert(
19    docs.map((doc, i) => ({
20      id: doc.id,
21      values: embeddings.data[i].embedding,
22      metadata: { text: doc.text, source: doc.source },
23    }))
24  );
25}
26 
27async function search(query: string, topK = 5) {
28  const { data: [{ embedding }] } = await openai.embeddings.create({
29    model: 'text-embedding-3-small',
30    input: [query],
31  });
32 
33  const results = await index.query({
34    vector: embedding,
35    topK,
36    includeMetadata: true,
37  });
38 
39  return results.matches.map(m => ({
40    score: m.score,
41    text: m.metadata?.text as string,
42    source: m.metadata?.source as string,
43  }));
44}
44 lines
Timeline — 4 Steps
Create Index
pc.index() connects to a Pinecone index — purpose-built vector store💡 Analogy
Like a specialized search engine — you don't build the search infrastructure, just use the API.
Pinecone is a managed vector database. Create an index with dimensions matching your embedding model (1536 for text-embedding-3-small). Connect to it with pc.index().
Console Output
> Pinecone client: authenticated
> Index "knowledge-base": 1536 dims, cosine metric
> Ready to upsert/query

Was this page helpful?

Discussion

pc.index() connects to a Pinecone index — purpose-built vector store

💡 Analogy

Like a specialized search engine — you don't build the search infrastructure, just use the API.

Pinecone is a managed vector database. Create an index with dimensions matching your embedding model (1536 for text-embedding-3-small). Connect to it with pc.index().

Console Output

> Pinecone client: authenticated
> Index "knowledge-base": 1536 dims, cosine metric
> Ready to upsert/query

pc.index() connects to a Pinecone index — purpose-built vector store

💡 Analogy

Like a specialized search engine — you don't build the search infrastructure, just use the API.

Pinecone is a managed vector database. Create an index with dimensions matching your embedding model (1536 for text-embedding-3-small). Connect to it with pc.index().

Console Output

> Pinecone client: authenticated
> Index "knowledge-base": 1536 dims, cosine metric
> Ready to upsert/query