Django Chain

A Django library for seamless LangChain integration, making it easy to add LLM capabilities to your Django applications.

This is a reusable Django application that provides a robust framework for defining, managing, and executing multi-step Large Language Model (LLM) workflows. It offers a set of API endpoints to interact with prompts, workflows, and their execution, enabling dynamic LLM applications without direct code changes for each new workflow. You can find the documentation here.

[!WARNING] This project is currently in early alpha and is still a work in progress. You could possibly experience specific builds failing to run, loss of data between upgrades and a lot of bugs. It is highly encouraged you wait for a stable release.

Features

• Dynamic prompt and workflow management out of the box
• Easy integration with existing Django models and views
• Built-in utilities for common LLM tasks
• Type-safe and well-documented API
• Comprehensive test coverage
• Customizable LLM logging and telemetry
• Production-ready with proper error handling

Core Concepts

• Prompt: Represents a configurable template for generating LLM prompts. This can be a HumanMessagePromptTemplate, SystemMessagePromptTemplate, or ChatPromptTemplate.
• Workflow: A sequence of ordered steps, each defining an action to be performed (e.g., format a prompt, call an LLM, parse JSON output, use a tool). Workflows orchestrate the flow of data through these steps.

Prerequisites

1. Python 3.9+
2. Django 4.0+

Installation

pip install django-chain

Add django_chain to your INSTALLED_APPS:

INSTALLED_APPS = [
    ...
    'django_chain',
    ...
]

Add you LLM model configurations:

DJANGO_LLM_SETTINGS = {
    "DEFAULT_LLM_PROVIDER": "fake",
    "DEFAULT_CHAT_MODEL": {
        "name": "fake-model",
        "temperature": 0.7,
        "max_tokens": 1024,
        "api_key": "fake key",
    },
    "DEFAULT_EMBEDDING_MODEL": {
        "provider": "fake",
        "name": "fake-embedding",
    },
    "VECTOR_STORE": {
        "TYPE": "pgvector",
        "PGVECTOR_COLLECTION_NAME": "test_documents",
    },
    "ENABLE_LLM_LOGGING": True,
    "LLM_LOGGING_LEVEL": "DEBUG",
    "MEMORY": {"PROVIDER": "django"},  # Use Django ChatHistory model as primary storage
    "CHAIN": {
        "DEFAULT_OUTPUT_PARSER": "str",
        "ENABLE_MEMORY": True,
    },
    "CACHE_LLM_RESPONSES": True,
    "CACHE_TTL_SECONDS": 3600,
}

Run migrations:

python manage.py makemigrations django_chain
python manage.py migrate django_chain

Quick Start

Add these urls to your app:

# your_project/urls.py
from django.contrib import admin
from django.urls import path, include

urlpatterns = [
    path('api/', include('django_chain.urls')), # Or your chosen app name
]

Development

1. Clone the repository
2. Install development dependencies:

   pip install -e ".[dev]"
   

3. Install pre-commit hooks:

   pre-commit install
   

4. Run tests:

   pytest
   

License

MIT License