Libro SDK

The official JavaScript/TypeScript SDK for Libro — The User Context Layer for AI Applications.

Transform conversations into structured, evolving user understanding. Build AI agents with infinite memory using our zero-cost local vectorization, semantic deduplication, and drop-in SDK.

Features

• ⚡️ Blazing Fast Ingestion: Edge-optimized architecture guarantees <300ms ingestion latency.
• 🧠 Zero Pipeline Changes: Seamlessly retrieve memories to pass into your LLM prompt.
• 🛡️ Full Privacy Control: Bring your own Supabase. We don't store your vectors, you do.
• ⏱️ Automated Context Trimming: Never overflow your context window again.

Installation

npm install libro-sdk

Note: Requires Node.js >= 18.

Quick Start

1. Initialize the Client

import { LibroClient } from 'libro-sdk';

const libro = new LibroClient({
  apiKey: process.env.LIBRO_API_KEY, 
  baseUrl: 'https://libro.co.in' // Optional: defaults to production
});

2. Ingest Memory

Save facts, preferences, or chunks of conversation into a user's memory brain.

const response = await libro.ingest({
  content: "User prefers functional programming over OOP.",
  endUserId: "user-12345",
  metadata: { source: "slack_integration" }
});

console.log(response.memory.id); // 'f3b974d0-...'

3. Retrieve Context

Retrieve the most relevant memories for a user before passing context to an LLM like GPT-4 or Claude.

const { context, memories } = await libro.getContext({
  query: "What kind of code structure do they like?",
  endUserId: "user-12345"
});

console.log(context); 
// -> "[Memory 1] (Relevance: 85.2%): User prefers functional programming over OOP."

4. Update Memory

When a user's preferences change, update their specific memory block directly.

await libro.update("memory-uuid-1234", {
  content: "User prefers OOP when building large enterprise systems."
});

5. Forget (GDPR Compliance)

Fully delete a user's memory or specific blocks to comply with data deletion requests.

// Delete a specific memory
await libro.forget({ memoryId: "memory-uuid-1234" });

// Delete ALL memories for a user (GDPR "Right to be Forgotten")
await libro.forget({ endUserId: "user-12345" });

How It Works Under The Hood

Unlike managed monoliths (Mem0, Zep), Libro uses an Edge-first architecture. Our SDK hits your API, and the vectorization happens instantly on the edge via a lightweight embedding sidecar (nomic-embed-text-v1.5), before being stored directly in your own Supabase instance using pgvector HNSW indexes.

Zero vendor lock-in. Full data sovereignty.

Libro — Memory that never forgets.