Skip to main content

Simple LLM Memory

Project description

Memento: Simple LLM Memory

Memento is a conversation management API for llm applications. It interfaces with your SQL database of choice to handle conversational histories.

Memento uses SQLAlchemy and Alembic under the hood to interact with SQL databases, so any database that is supported by these libraries (PostgreSQL, MySQL, SQLite, CosmoDB, etc.) is also supported by Memento.

Installation

$ pip install memento-llm

Getting Started

With Memento, you no longer have to worry about setting up message storage logic in your application, allowing for a seamlessly stateless flow, here is how it can be integrated into your code:

Recorder API

Currently Memento only has the Recorder API, which serves as a simple way to use Memento in applications dependent on SQLAlchemy sessions. Because of this fact, it is a natural fit for FastAPI applications (which is my main use for Memento, personally).

The main differentiator of the Recorder API is that it requires that a SQLAlchemy Session or AsyncSession be provided. The same base Recorder class has methods to use both types of sessions.

from openai import OpenAI
from memento import Recorder, crud, models
from sqlalchemy import create_engine
from sqlalchemy.orm import Session

# Setup

client = OpenAI()
engine = create_engine("sqlite://") # In-memory sqlite database

models.Base.metadata.create_all(engine) # For demo purposes, create tables with the metadata API

with Session(engine) as session:
  conversation_id = crud.create_conversation(session, "Testbot") # Name of the assistant/agent/app

# Usage

def generate():
  # Start the recorder with previous conversation data (Empty the during the first call, one message during the second)
  recorder = Recorder.from_conversation(session, conversation_id)

  # Call the LLM API with data retrieved from the recorder
  response = client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages=recorder.to_openai_format()
  )

  # Add the response to the recorder
  recorder.add_openai_response(response)

  # Commit new messages to your database
  recorder.commit_new_messages(session)

response_1 = generate("My name is Anibal")
print(response_1) # Output: Hello Anibal!

response_2 = generate("What´s my name?")
print(response_2) # Output: Your name is Anibal.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

memento_llm-0.2.93.tar.gz (5.3 kB view details)

Uploaded Source

Built Distribution

memento_llm-0.2.93-py3-none-any.whl (6.6 kB view details)

Uploaded Python 3

File details

Details for the file memento_llm-0.2.93.tar.gz.

File metadata

  • Download URL: memento_llm-0.2.93.tar.gz
  • Upload date:
  • Size: 5.3 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/1.8.3 CPython/3.12.3 Linux/6.8.0-48-generic

File hashes

Hashes for memento_llm-0.2.93.tar.gz
Algorithm Hash digest
SHA256 68035d8a6697cc6ab43c236f9cb211bb4a41eb258364a481adf006fa4e2db797
MD5 3ac47885c756df6922562ff4ee109dc7
BLAKE2b-256 ca2724adad5558e4a8b4f623844c2d188a60509dce0fd3a739e149d19fad63e7

See more details on using hashes here.

File details

Details for the file memento_llm-0.2.93-py3-none-any.whl.

File metadata

  • Download URL: memento_llm-0.2.93-py3-none-any.whl
  • Upload date:
  • Size: 6.6 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/1.8.3 CPython/3.12.3 Linux/6.8.0-48-generic

File hashes

Hashes for memento_llm-0.2.93-py3-none-any.whl
Algorithm Hash digest
SHA256 bfde002bcde8197f9123f75f220b2f11ade36a7128234be106356e29b3ff9754
MD5 acde85c586874412f62d5b37da11ae85
BLAKE2b-256 3d7221af4f120adc0ba69364e40f76312aed705eb919b1029d16d6d041f1d278

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page