iragent 0.1.8

A simple multi-agent framework

iragent

iragent is a simple framework for building OpenAI‑Like, tool‑using software agents.
It sits halfway between a prompt‑engineering playground and a full orchestration layer—perfect for experiments, research helpers and production micro‑agents.

---

✨ Key features

Feature	Why it matters
Composable Agent model	Chain or orchestrate agents via SimpleSequentialAgents, AgentManager, and AutoAgentManager for flexible workflows
Auto-routing agent	AutoAgentManager uses a language model to dynamically decide the next agent in the loop
Web-augmented agent	InternetAgent uses googlesearch, requests, and summarizing agents to fetch and condense live web data
Parallel summarization	fast_start method uses ThreadPoolExecutor to speed up web content processing
Prompt-driven summaries	Summarization is driven by customizable system prompts and token-limited chunking for accurate context
Simple, Pythonic design	Agents are lightweight Python classes with callable message interfaces—no metaclasses or hidden magic
Memory, BaseMemory	BaseMemory provides foundational memory management for agents, storing conversation history and message objects. It supports adding, retrieving, and clearing memory, offering a flexible design for session-based context, interaction history, or task-specific memory across multiple agent invocations. Ideal for scenarios where the agent needs to recall past interactions for continuity.
SummarizerMemory with summarizer agent	SummarizerMemory extends BaseMemory by integrating a summarizing Agent that automatically condenses long histories when memory limits are exceeded. This enables agents to maintain compact, relevant context over time, ensuring efficiency without losing key information.
SmartAgentBuilder for automated agent creation	SmartAgentBuilder automates breaking down a high-level task into structured subtasks, then creates specialized agents for each subtask using a sequential pipeline. It ensures that each agent is precisely configured with a strict role, and outputs an AutoAgentManager to run them in coordination.

---

SimpleAgenticRAG + KnowledgeGraphBuilder

SimpleAgenticRAG combines a FAISS-powered retriever (KnowledgeGraphBuilder) with agent-based orchestration for question answering.
It follows a Retriever → Generator flow: search relevant chunks, then generate an answer with your LLM.

Example (Local LLM)

Installation:

pip install iragent[rag]

from sentence_transformers import SentenceTransformer
from iragent.models import KnowledgeGraphBuilder
from iragent.agent import AgentFactory
from iragent.models import SimpleAgenticRAG

base_url= "http://127.0.0.1:1234/v1" # use your own base_url from api provider or local provider like ollama.
api_key = "no-key" # use your own api_key.
model = "qwen3-4b-instruct-2507"

emb_model = SentenceTransformer("all-MiniLM-L6-v2")
kg = KnowledgeGraphBuilder(embedding_model=emb_model, index_dir="./text-store/")
texts = ["FAISS (Facebook AI Similarity Search) is a library for efficient similarity search and clustering of dense vectors.",
        "It is written in C++ with bindings for Python, and is widely used for large-scale nearest-neighbor search.", 
        "FAISS supports both exact search and approximate search algorithms, making it flexible for different speed/accuracy needs.", 
        "The library was developed by Facebook’s AI Research (FAIR) team."]
kg.build_index_from_texts(texts)
agent_factory = AgentFactory(
    base_url=base_url,
    api_key=api_key,
    model=model,
)

rag = SimpleAgenticRAG(
    kg = kg,
    agent_factory= agent_factory
)
answer = rag.ask("Who developed FAISS?")

See examples/SimpleAgenticRAG for more usage.

🚀 Installation

# Requires Python 3.10+
pip install iragent
# For AgenticRAG
pip install iragent[rag]
# For AgenticRAG with GPU
pip install iragent[rag-gpu]
# Or directly from GitHub
pip install git+https://github.com/parssky/iragent.git

⚡ Quick start

from iragent.tools import get_time_now, simple_termination

factory = AgentFactory(base_url,api_key, model)

agent1 = factory.create_agent(name="time_reader",
                            system_prompt="You are that one who can read time. there is a fucntion named get_time_now(), you can call it whether user ask about time or date.",
                            fn=[get_time_now]
                            )
agent2 = factory.create_agent(name="date_exctractor", 
                              system_prompt= "You are that one who extract time from date. only return time.")
agent3 = factory.create_agent(name="date_converter", 
                              system_prompt= "You are that one who write the time in Persian. when you wrote time, then in new line write [#finish#]")

manager = AutoAgentManager(
    init_message="what time is it?",
    agents= [agent1,agent2,agent3],
    first_agent=agent1,
    max_round=5,
    termination_fn=simple_termination,
    termination_word="[#finish#]"
)

res = manager.start()
res.content

More docs

visit below url: https://parssky.github.io/iragent/namespacemembers.html

📚 More Usage Examples

Explore practical examples and use cases in the example directory.

Development

git clone https://github.com/parssky/iragent.git
cd iragent
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"          # ruff, pytest, etc.

🤝 Contributing

Pull requests are welcome! Please open an issue first if you plan large‑scale changes. 1- Fork → create feature branch

2- Write tests & follow ruff style (ruff check . --fix)

3- Submit PR; GitHub Actions will run lint & tests.

📄 License

This project is released under the MIT License.