llamate 0.1.7

A memory-augmented framework for LLMs

🦙 Llamate

Llamate is a memory-augmented agent framework for Large Language Models (LLMs) that provides persistent, retrievable memory for AI conversations.

What is Llamate?

Llamate solves a fundamental limitation of current LLMs: their inability to remember past conversations beyond a single context window. It creates a vector database of memories that can be semantically searched and retrieved during conversations, allowing LLMs to maintain continuity and context over extended interactions.

How It Works

1. Memory Storage: Llamate stores important pieces of conversation as vector embeddings in a database (either FAISS or PostgreSQL).
2. Semantic Retrieval: When new queries come in, Llamate searches for semantically relevant past memories.
3. Memory Filtering: The system automatically filters out the current query from search results to prevent echo effects.
4. Context Enhancement: Retrieved memories are injected into the conversation context, allowing the LLM to access and utilize past information.
5. User Identification: Each user gets a unique memory space, ensuring personalized conversation history.

Key Features

• Multiple Backend Support: Works with FAISS (file-based) or PostgreSQL (with pgvector)
• Persistence: Memories remain available between sessions and application restarts
• Simple API: Easy-to-use Python interface that works with any LLM
• CLI Interface: Command-line tool for quick testing and interaction
• Production Ready: Designed for both development and production environments

Quick Start

Follow these steps to set up, use, and view data in Llamate with PostgreSQL:

1. Install Llamate

pip install llamate

2. OpenAI API Requirements

Llamate requires access to the following OpenAI models in your account:

• text-embedding-3-large - Used for vector embeddings
• gpt-4 - Recommended for high-quality responses

Make sure these models are enabled in your OpenAI account and set your API key:

3. Start PostgreSQL Container

docker run --name llamate-postgres -e POSTGRES_USER=llamate -e POSTGRES_PASSWORD=llamate -e POSTGRES_DB=llamate -p 5432:5432 -d ankane/pgvector

4. Initialize Llamate

llamate --init
# Select 'postgres' as your vector store backend
# Enter connection string: postgresql://llamate:llamate@localhost:5432/llamate

5. Test Llamate locally

from llamate import MemoryAgent, get_vectorstore_from_env

# Set user ID
user_id = "test_user"

# Initialize components
vectorstore = get_vectorstore_from_env(user_id=user_id)
agent = MemoryAgent(user_id=user_id, vectorstore=vectorstore)

# Add memories
agent.chat("The capital of France is Paris.")
agent.chat("The Eiffel Tower is 324 meters tall.")
agent.chat("Python is a programming language created by Guido van Rossum.")

# Test retrieval
response = agent.chat("Tell me about Paris.")
print("Response:", response)

6. View Data in PostgreSQL

Connect to the database:

docker exec -it llamate-postgres psql -U llamate -d llamate

List tables to find your memory table (it will use your user_id):

\dt

View table structure:

\d memory_test_user

Display memory records (omitting the large vector field):

SELECT id, text FROM memory_test_user;

Count records:

SELECT COUNT(*) FROM memory_test_user;

Query specific memories (using text search):

SELECT id, text FROM memory_test_user WHERE text LIKE '%Paris%';

Delete test memories (if needed):

DELETE FROM memory_test_user WHERE text LIKE '%test%';

Exit the PostgreSQL shell:

\q

Features

• Persistent memory for AI using vector embeddings
• Multiple vector store backends (FAISS and PostgreSQL)
• Easy integration into existing applications
• Simple CLI for testing and demonstration