identt 0.1.20

Python SDK for Iden API interactions

Iden

Introduction

Iden is an advanced framework designed to simplify the development and deployment of LLM-based (Large Language Model) applications. Whether you’re working on Retrieval-Augmented Generation (RAG), building sophisticated chatbots, or generating AI-driven insights, Iden streamlines the process with an intuitive and efficient pipeline.

Iden empowers users to define models enriched with system prompts and vector-based document indexes, seamlessly supporting formats like .txt and .json. These indexes and models can be quickly configured, enabling the rapid creation of AI applications, significantly reducing development time.

With features like token tracking for AI responses and support for dynamic prompts similar to OpenAI's chat completion—Iden provides robust tools for building scalable AI solutions tailored to diverse use cases.

Identt SDK allows you to interact with Iden thorough an easy to use client interface to implement your AI applications.

Installation

Identt SDK is available on PyPI and can be installed using pip. To install the latest version of Iden, run the following command:

pip install identt

Also you can add Iden to your .pipfile under [packages] when deploying:

identt = "*" 

or

identt = "<version>" 

Prerequisites

Before using the Identt SDK, ensure you have the following ready:

---

1. Backend Base URL

You need the base URL of the backend system that the Identt client will interact with. This URL will be provided by the admin.

---

2. User Token

A user token is required for authentication when interacting with the Identt backend. To obtain this token:

• Request Admin:
  • Contact your admin or project manager to create a user or a project-specific account for you.
  • The admin will provide an associated token upon creating the user.

---

3. Initiating the Client

Once you have the Base URL and Token, you can initialize the Identt client as follows:

from identt import client

# Replace BASE_URL and TOKEN with your actual backend URL and user token
client = client.Client(base_url=BASE_URL, token=TOKEN)

Examples

The following examples demonstrate the basic workflow and showcases sample use cases to help you get started quickly.

Example 1: Simple RAG based Chatbot Application

1. Read Constants from .env file:

It is a best practice to store BASE_URL and TOKEN values in a .env file and read them dynamically in your code.

import os

BASE_URL = os.getenv("IDEN_BASE_URL")
TOKEN = os.getenv("IDEN_USER_TOKEN")

2. Initialize the Client:

from identt import client

client = client.Client(base_url=BASE_URL, token=TOKEN)

3. Create the Model:

First you need to define a SYSTEM_PROMPT to create the model.

SYSTEM_PROMPT = "You are a helpful assistant."

Then you can define a unique model_name and create the model as below.

model_name = "chatbot_model"
model = client.create_model(name=model_name, system_prompt=SYSTEM_PROMPT)
model_id = client.get_model_id_by_name(name=model_name)

4. Index Files:

First create a dictionary of file objects to be indexed in this RAG based chatbot application, as below.

file_objects = {}

file_objects["data-1.json"] = open("data/data-1.json", "rb")
file_objects["data-2.json"] = open("data/data-2.json", "rb")
file_objects["data-3.txt"] = open("data/data-3.txt", "rb")

---

Note: Iden only supports .json, .txt and .rtf currently. You need to convert .pdf,.doc or any other format to one of the supported fromats above.

---

Then pass this dictionary of file objects to a new index with unique name defined as below.

index_name = "chatbot_index"
index = client.index_files(file_objects=file_objects, name=index_name)
index_uuid = client.get_index_uuid_by_name(name=index_name)

5. Perform Chat:

Finally perform the chat by passing model_id and index_uuid along with a query/question of your choice.

 response = client.chat(
        model_id=model_id,
        index_uuid=index_uuid,
        query="Tell me everything you know about Danu Kumanan",
    )

Response:

{
  'response': "Danu Kumanan is the CTO and Co-founder of 1 Club. He demonstrated entrepreneurial skills early on by selling his first website at the age of 13. He holds a Master's degree in AI & Mathematics and has developed his expertise as a blockchain engineer at prominent companies such as Dapper Labs, Flow, and Coinbase. His career highlights his dedication to advanced technology and leadership within the blockchain sector.", 
  'tokens': 390
}

6. Further Usage

The created indexes and models are persistently stored in Iden backend. So no need to recreate or reindex files when using again. You can get the created model_id and index_uuid by its unique name as below for further use.

model_id = client.get_model_id_by_name(name="chatbot_model")
index_uuid = client.get_index_uuid_by_name(name="chatbot_index")

You can add more files and update your created index as below.

new_file_objects = {"data-4.txt": open("data/data-4.txt", "rb")}

index_uuid = client.get_index_uuid_by_name(name="chatbot_index")
index = client.index_files(file_objects=new_file_objects, index_uuid=index_uuid)

Example 2: AI Application with Dynamic Prompts

One of the limitations of above example is that the SYSTEM_PROMT of the created model cannot be changed once created. Further some AI applications operate solely based on user prompts and do not require file indexing, providing responses directly from the input prompt which are subjected to change. For such use cases we can simply use basic chat functionality which works quite similar to chat completion API of OpenAI.

DYNAMIC_PROMPT = f'dynamically changing prompt here: {prompt}'
PROMPT = DYNAMIC_PROMPT.fomrat(prompt=user_prompt)

response = client.basic_chat(prompt=PROMPT, gpt_model="gpt-4o-mini")

---

Note: If gpt_model parameter is not specified Iden will by default use gpt-4o model. The gpt_model parameter should be one of the below:

• gpt-3
• gpt-3.5-turbo
• gpt-4
• gpt-4-turbo
• gpt-4-mini
• gpt-4o
• gpt-40-mini
• gpt-4o-2024-08-06
• o1-preview
• o1-mini

---