libro-sdk 0.1.1

Official Python SDK for the Libro AI memory infrastructure.

Libro Python SDK

The official Python 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 LangChain or LlamaIndex prompts.
• 🛡️ Full Privacy Control: Bring your own Supabase. We don't store your vectors, you do.
• ⏱️ Automated Context Trimming: Never overflow your LLM context window again.

📦 Installation

pip install libro-sdk

🚀 Quick Start

1. Initialize the Client

from libro import LibroClient
import os

# Initialize the client with your API key
client = LibroClient(
    api_key=os.environ.get("LIBRO_API_KEY"), 
    base_url="https://libro.co.in" # Optional: defaults to production
)

2. Ingest Memory

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

response = client.ingest(
    user_id="user-12345",
    text="User prefers Python over Java for machine learning.",
    metadata={"source": "slack_integration"}
)

print(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.

context_data = client.get_context(
    user_id="user-12345",
    query="What programming language should I use for their new ML project?"
)

print(context_data["context"]) 
# -> "[Memory 1] (Relevance: 85.2%): User prefers Python over Java for machine learning."

4. Update Memory

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

client.update(
    memory_id="memory-uuid-1234",
    content="User prefers Python for ML, but uses Rust for high-performance systems."
)

5. Forget (GDPR Compliance)

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

# Delete a specific memory
client.forget(memory_id="memory-uuid-1234")

# Delete ALL memories for a user (GDPR "Right to be Forgotten")
client.forget(user_id="user-12345")

🏗️ How It Works Under The Hood

Unlike managed monoliths, 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.