Name: Cencori
Rating: 5 (1 reviews)

Neon gives you Serverless Postgres with branching. Cencori gives you Serverless AI with memory.

This combination is powerful because it allows you to separate Structured Data (Neon) from Unstructured Memory (Cencori).

The Architecture

Neon: Stores Users, Orders, Products, and Transactional Data.
Cencori: Stores Embeddings, Chat History, User Preferences, and Observations.

Connecting from Vercel / Node.js

You don't connect Neon directly to Cencori. Instead, your application queries both.

Codetext

import { neon } from '@neondatabase/serverless';
import { Cencori } from 'cencori';
 
const sql = neon(process.env.DATABASE_URL);
const cencori = new Cencori();
 
async function handleRequest(query: string) {
    // 1. Get structured context from Neon
    const products = await sql`SELECT name, price FROM products WHERE stock > 0`;
 
    // 2. Get unstructured context from Cencori (RAG)
    const knowledge = await cencori.memory.search(query);
 
    // 3. Synthesize answer
    const response = await cencori.chat.create({
        model: "gpt-4o",
        messages: [
            { role: "system", content: `You are a sales assistant.` },
            { role: "system", content: `Available Products: ${JSON.stringify(products)}` },
            { role: "system", content: `Relevant Policy: ${JSON.stringify(knowledge)}` },
            { role: "user", content: query }
        ]
    });
 
    return response;
}

Tool Calling (SQL Agent)

You can give Cencori the ability to query Neon directly using Tools.

Warning: Always use read-only credentials or strict input validation when giving an LLM access to SQL.

Codetext

const tools = [
  {
    type: "function",
    function: {
      name: "search_users",
      description: "Search for a user by email",
      parameters: {
        type: "object",
        properties: {
          email: { type: "string" }
        },
        required: ["email"]
      }
    }
  }
];
 
// ... inside your chat loop
if (toolCall.name === 'search_users') {
    const { email } = JSON.parse(toolCall.arguments);
    const user = await sql`SELECT * FROM users WHERE email = ${email}`;
    // ... return user to LLM
}