Agentiq2 is an easy-to-use AI agent framework developed using Python and uses OpenAI's API (The previous version agentiq used Cohere's API and models). It has minimal dependencies and does not rely on LangChain or other agent frameworks. This repo consists of 2 parts - the Core agent framework and a Web App for demonstrating the functionality.

Core

Agents interact with their environment and perform actions using Tools. In Agentiq, tools are nothing but Python functions. Consider the below function that creates an event in a user's calendar.

def create_event(
    name: Annotated[str, "Name of the event"],
    date: Annotated[str, "Date of the event, formatted as YYYY-MM-DD"],
    start_time: Annotated[str, "Start time of the event, formatted as HH:MM"],
    end_time: Annotated[str, "End time of the event, formatted as HH:MM"],
):
    """Creates a new event based on event name, date, start and end time"""
    new_event = {"name": name, "date": date, "start_time": start_time, "end_time": end_time}
    events.append(new_event)
    return f"Event '{name}' created successfully for {date} from {start_time} to {end_time}"

This can be converted into a tool with the following code:

create_event_tool = Tool(create_event)

An Agent can then be created by providing a list of tools that it can utilize.

event_agent = Agent([create_event_tool])

A system message can also be provided to the agent to give it more context. The system message can either be a string or a function including a lambda.

tools = [create_event_tool]
system_message = lambda: f"Today is {date.today()}"
event_agent = Agent(tools=tools, system_message=system_message)

The agent can be invoked using its run method.

generator = event_agent.run(message="Create an hour-long appointment for the first available free slot after 9am titled Client Meeting")
for response in generator:
    print(response)

Web App

The Web App is a basic CRM system that demonstrates the Agentiq framework. The following outlines the various functionalities of the web app and the tools in each.

Customers

The Customers section displays all customer records. Here are the tools available for managing customer information:

Tool	Example Prompts
get_customer_by_name: Returns customer details including address based on the name	"Where is TechNova located?"
create_customer: Creates a customer record by providing the customer's name, city, state and country	"Create a customer for Presalesly AI located in Atlanta, Georgia, United States"
edit_customer: Edits a customer record given the customer ID and new details	"Update the city for GreenEdge Innovations to Palo Alto"

Opportunities

In the Opportunities section, you can view all the opportunities in CRM. Available tools include:

Tool	Example Prompts
get_opportunity_by_name: Returns opportunity details based on the opportunity name	"What is the value and stage of the Cloud Migration Project opportunity?"
create_opportunity: Creates an opportunity record by specifying the opportunity name, customer ID and amount	"Create an opportunity called HCM Implementation for Skyline Dynamics for 120000"
close_opportunity: Closes an opportunity given the opportunity name and stage	"Close the Cybersecurity Upgrade opportunity"
edit_opportunity: Edits an opportunity record given the opportunity ID and new details	"Update the CRM Expansion opportunity to 85K"
delete_opportunity: Deletes an opportunity record given the opportunity ID	"Delete the Network Security Audit opportunity"

Events

The Events section displays all events for a selected date (default is today). Here are the tools available for managing events:

Tool	Example Prompts
get_events_by_date: Returns all events for a specific date	"What meetings do I have today?"
create_event: Creates a new event based on event name, date, start and end time	"Create an hour-long appointment for the first available free slot after 9am today titled RFP response review"

Tech Stack

The Core is developed using Python and uses OpenAI's API which means that it can be used with any compatible inference provider. The Web App is developed using Python and FastAPI as the backend and Vue (written in TypeScript) as the frontend. It is hosted on Render at agentiq2.onrender.com where each commit & push to github triggers a deploy.