servifai 0.2.1

A mini framework built on top of Langchain and Llamaindex to provide LLM powered Autonomous Agents as a simplified service to assist users with their tasks

ServifAI

A mini framework built on top of Langchain and Llamaindex to provide LLM powered Autonomous Agents as a simplified service to assist users with their tasks.

Overview

ServifAI (Task-based Agent) = LLM + Memory + Planning + Toolbox

Instead of feeding all kinds of tools to a single agent and confusing it while selection, ServifAI narrows down the selection by combining only necessary tools on basis of the task at hand.

Read this article to get an overview on Agents.

Current Supported Tasks:

Tasks	Toolbox Tools
default	DuckDuckGo + LLM Math + PAL Math
qna_local_docs	Vector Index + Knowledge Graphs

Installation

Works best with Poetry

poetry add servifai

With pip, you might have to install dependencies manually

pip install servifai

Usage

Create a .env file

OPENAI_API_KEY='sk-...'

Run Python Code

from servifai import ServifAI
myagent = ServifAI()

while True:
    text_input = input("Me: ")
    if text_input == "exit":
        break
    response = myagent.chat(text_input)
    print(f'ServifAI: {response}\n')

Output

Me: Hi, How are you?
ServifAI: Hi, I'm an AI language model, so I don't have feelings, but I'm here to help you with any questions or tasks you have!

Me: What is the current weather of Bengaluru?
ServifAI: The current weather in Bengaluru is mostly cloudy with a temperature of 81°F (27°C). The wind is coming from the north at 3 mph (5 km/h). Tomorrow's temperature is expected to be nearly the same as today.

Data Creation Recipe for Local Docs Knowledge Extraction Tasks

Consider the example of Uber 10Q filings.

• Download the quaterly reports for year 2022 and 2023 as pdf and save it locally in a directory (here reports).
• Create a config YAML file uber10q.yaml inside a configs dir and fill details as:

task: qna_local_docs

llm:
  org: openai
  model: gpt-3.5-turbo
  temperature: 0
  max_tokens: 3000

data:
  dir: reports
  about: "Uber 10Q Filing"

memory:
  dir: uber_10q
  max_input_size: 2500
  num_outputs: 1000
  max_chunk_overlap: 0.05
  chunk_size_limit: 1000

• As the task is to extract information from these pdfs, so task chosen is qna_local_docs. Based on these tasks ServifAI chooses toolbox which contains specific tools required for task completion. We will be adding more toolbox later.
• To achieve optimum results, its recommended to rename your pdfs as few words description. For example, we rename quarterly 10Q reports as Q1-23.pdf, Q4-22.pdf etc. Do not add blank spaces between words, instead use hyphen -.
• Also in config file in data.about, provide a concise common summary of all these multiple pdfs. For example, here we write this as Uber 10Q Filing.
• Run Python Code

from servifai import ServifAI
myagent = ServifAI('configs/uber10q.yaml')

while True:
    text_input = input("Me: ")
    if text_input == "exit":
        break
    response = myagent.chat(text_input)
    print(f'ServifAI: {response}\n')

• Output

Me: Analyze the revenue growth of Uber across last few quarters
ServifAI: Based on the provided context, Uber's revenue growth in the last few quarters can be summarized as follows:

- Q3 2022: 72% year-over-year or 81% on a constant currency basis.
- Q4 2022: 49% year-over-year.
- Q1 2023: 29% year-over-year or 33% on a constant currency basis.

Therefore, Uber's revenue growth in the last few quarters has been positive, with varying rates of growth.

Me: How much cash did Uber have in last quarter of 2022?
ServifAI: Based on the provided context, the cash balance for Uber in Q4 2022 was $4.3 billion.

TODO:

• Add other task based tools
• Add support for Local LLMs
• Add support for other VectorDBs
• Add support for other unstructured data
• Add support for structured data
• OpenAI funcs