UnisonAI is a flexible and extensible Python framework built on agent to agent (a2a) for building, coordinating, and scaling multiple AI agents—each powered by the LLM of your choice, unisonai helps in making individual agents on focused tasks as well as clan-based agent for extensive and complex tasks.

Table of Content:

• Overview
• Why UnisonAI?
• Installation
• Core Components
• Parameter Reference Tables
• Usage Examples
• Faq?
• Contributing And License

Overview

UnisonAI is a flexible and extensible Python framework for building, coordinating, and scaling multiple AI agents—each powered by the LLM of your choice.

• Single_Agent: For solo, focused tasks.
• Agent (as part of a Clan): For teamwork, coordination, and distributed problem-solving.
• Tool System: Easily augment agents with custom, pluggable tools (web search, time, APIs, your own logic).

Supports Cohere, Mixtral, Groq, Gemini, Grok, OpenAI, Anthropic, and any custom model (just extend BaseLLM). UnisonAI is designed for real-world, production-grade multi-agent AI applications.

---

Uses A2A (Agent to Agent) Communication!

---

Why UnisonAI?

🔗 Multi-LLM Support Mix and match LLM providers with ease.	🧩 Modular & Extensible Add your own tools, LLMs, and logic.	🤖 Single or Multi-Agent Solo agents or collaborative Clans—your call. With Agent to Agent communication on your supervision
🛡️ Robust Error Handling Built-in JSON/YAML repair & retries.	📚 Clear Docs & Examples Well-documented APIs and quick starts.	⚡ Production-Ready Designed for real-world automation & chatbots.

---

Installation

Requires Python >=3.10, <3.13

pip install unisonai
# or
pip3 install unisonai

---

Core Components

Component	Purpose	Highlights
Single_Agent	Standalone agent for independent tasks	Own log/history Plug in any LLM/tools
Agent	Works with others in a Clan (team)	Inter-agent messaging Tools & role-based prompts
Clan	Manages a team of Agents (including a leader/manager)	Coordinated planning Shared instructions/goals
Tool System	Extend agent capabilities with custom logic	Reusable, pluggable tools Strongly-typed parameters

---

Parameter Reference Tables

Single_Agent

Parameter	Type	Description	Default
llm	BaseLLM/any LLM	LLM for the agent	Required
identity	String	Agent's unique name	Required
description	String	Agent’s purpose	Required
verbose	Boolean	Verbose/debug mode	True
tools	List	Usable tools	[]
output_file	String	Output file path	None
history_folder	String	Directory for logs/history	"history"

Agent

Parameter	Type	Description	Default
llm	Gemini/any LLM	LLM for the agent	Required
identity	String	Agent's unique name	Required
description	String	Responsibilities overview	Required
task	String	Agent’s core goal/task	Required
verbose	Boolean	Verbose/debug mode	True
tools	List	Usable tools	[]

Clan

Parameter	Type	Description	Default
clan_name	String	Name of the clan	Required
manager	Agent	Clan manager/leader	Required
members	List	List of Agent instances	Required
shared_instruction	String	Instructions for all agents	Required
goal	String	Unified clan objective	Required
history_folder	String	Log/history folder	"history"
output_file	String	Final output file	None

BaseTool & Field

BaseTool

Attribute/Method	Description
name	Tool name
description	Tool function summary
params	List of Field objects (inputs)
_run(**kwargs)	Tool logic implementation

Field

Attribute	Description	Default
name	Parameter name	Required
description	Parameter purpose	Required
default_value	Default value	None
required	Is parameter mandatory?	True

---

Usage Examples

Standalone Agent

This is the code from main.py FILE.

from unisonai import Single_Agent
from unisonai.llms import Gemini
from unisonai.tools.websearch import WebSearchTool

web_agent = Single_Agent(
    llm=Gemini(model="gemini-2.0-flash"),
    identity="Web Explorer",
    description="Web Searcher for multiple queries",
    tools=[WebSearchTool],
    history_folder="history",
    output_file="output.txt"
)

web_agent.unleash(task="Find out what is the age of Trump")

---

Clan-Based Agents

"This is a refernce from main2.py FILE, check the file for the full complex example."

from unisonai import Agent, Clan
from unisonai.llms import Gemini
from unisonai.tools.websearch import WebSearchTool

time_agent = Agent(
    llm=Gemini(model="gemini-2.0-flash"),
    identity="Time Keeper",
    description="Handles scheduling",
    task="Track time-related info for the trip",
    tools=[TimeTool]
)

web_agent = Agent(
    llm=Gemini(model="gemini-2.0-flash"),
    identity="Web Explorer",
    description="Searches for travel info",
    task="Gather travel info, cultural activities, and costs",
    tools=[WebSearchTool]
)

clan = Clan(
    clan_name="Ultimate Trip Expert Clan",
    manager=planner_agent,  # Your clan leader agent
    members=[time_agent, web_agent /*, ...other agents*/],
    shared_instruction="Collaborate to plan a budget-friendly 7-day trip in India.",
    goal="Plan a 7-day itinerary across multiple cities with a budget of 10,000 INR",
    history_folder="trip_history",
    output_file="trip_plan.txt"
)

clan.unleash()

---

FAQ?

What is UnisonAI?

A Python framework for orchestrating multiple AI agents—each powered by your choice of LLMs, working solo or in teams. Specifically with A2A communication.

Why use a Clan?

For complex, multi-step, or specialized tasks: divide and conquer with specialized agents, coordinated by a manager.

Can I add my own LLM?

Yes! Just extend the BaseLLM class and plug in your model.

What are tools?

Reusable logic/components, built on BaseTool, that agents can call for specialized tasks (e.g., web search, APIs).

How is agent history logged?

Each agent maintains its own logs/history in the specified directory (default: history/).

What can I build with UnisonAI?

Chatbots, collaborative planners, research assistants, workflow automation, and more!

---

Contributing and License

Founder: @E5Anant
PRs and issues welcome! The project is under the Apache 2.0 License.

Open issues • Submit PRs • Suggest improvements

---

UnisonAI: Orchestrate the Future of Multi-Agent AI.