memloop 0.2.0

A local-first, dual-memory engine for AI Agents.

MemLoop

The Plug-and-Play Memory Engine for AI Agents.

MemLoop is a local-first Python library that gives LLMs "Infinite Memory". It ingests documents (PDF, CSV, TXT) and websites, stores them in a local vector database, and retrieves them with citation-style mapping.

No API keys required. 100% Offline Capable.

---

Quick Start

Installation

pip install memloop

The 30-Second Demo (Interactive CLI)

memloop

---

Build Your First RAG Agent (The 20-Line Tutorial)

Retrieval-Augmented Generation (RAG) usually requires setting up Vector DBs, Embedding Models, and Retrievers. MemLoop handles that complexity so you can focus on the logic.

With Gemini

import google.generativeai as genai
from memloop import MemLoop

# --- Configuration ---
genai.configure(api_key="YOUR_GEMINI_API_KEY")
model = genai.GenerativeModel('gemini-pro')
brain = MemLoop()

# --- Step 1: Ingest (Run once, persist forever) ---
# brain.learn_url("https://docs.python.org/3/glossary.html") 

# --- Step 2: Retrieve ---
query = "What is a decorator in Python?"
context = brain.recall(query) # <--- MemLoop does the heavy lifting

# --- Step 3: Generate ---
prompt = f"Use this context to answer:\n{context}\n\nUser: {query}"
response = model.generate_content(prompt)

print(response.text)

With OpenAI

from openai import OpenAI
from memloop import MemLoop

client = OpenAI(api_key="YOUR_KEY")
brain = MemLoop()

# brain.learn_local("./my_docs")

query = "Summarize the documents."
context = brain.recall(query)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": f"Context:\n{context}\n\nQ: {query}"}]
)
print(response.choices[0].message.content)

---

API Reference

MemLoop()

The main entry point. Initializes the local vector store (ChromaDB) in ./memloop_data.

brain = MemLoop(db_path="./custom_folder")

.learn_url(url: str)

Scrapes a webpage, cleans the HTML, chunks the text, and stores vectors locally.

• Returns: int (Number of chunks ingested).

.learn_local(folder_path: str)

Recursively ingests a local folder. Supports .pdf (with page tracking), .csv (row linearization), .txt, and .md.

• Returns: int (Number of documents processed).

.recall(query: str)

Retrieves the most relevant context.

1. Checks Semantic Cache (O(1) return if query is repeated).
2. If miss, performs Vector Search (Cosine Similarity).
3. Returns formatted string with Citations.

---

Pro Tip: The "Persistent Brain"

Because MemLoop uses ChromaDB locally, you don't need to run .learn_url every time!

Run the script once to learn:

brain.learn_url("https://docs.python.org") # Run this ONCE

Then comment it out. Your agent now "remembers" that data forever in future runs.

---

Technology Stack

• Vector Querying: ChromaDB
• Embeddings: all-MiniLM-L6-v2 (HuggingFace)
• Parsing: BeautifulSoup4 (Web), PyPDF (Docs)

---

Built with ❤️ by Vansh.