micro-agent-router

A zero-dependency, ultra-lightweight state machine for building agentic AI workflows. It gives you the routing power of LangGraph, without the steep learning curve and massive framework bloat.

The Problem

Building Agentic workflows (like a Researcher LLM passing data to a Writer LLM) requires a state machine. Frameworks like LangGraph or AutoGen are powerful but incredibly heavy. They require you to learn custom decorators, specific graph compiling steps, and complex state management classes.

Sometimes, you just want to run standard Python functions in a loop based on their outputs.

The Solution

micro-agent-router is a pure Python library that acts as a simple traffic controller. You register standard Python functions as "nodes". Each function just returns a dictionary telling the router where to go next.

Installation

pip install micro-agent-router

Usage

Here is a full agentic workflow (Writer -> Reviewer) in just a few lines of vanilla Python:

from micro_agent_router import MicroAgentRouter

# 1. Define your standard Python functions (Nodes)
def writer_agent(state: dict):
    print("Writing article...")
    state["content"] = "This is a great article."
    
    # Just return where to route next!
    return {"next": "reviewer_agent", "data": state}

def reviewer_agent(state: dict):
    print("Reviewing article...")
    # Add your LLM logic here to decide if it's good or bad
    is_good = True 
    
    if is_good:
        state["status"] = "published"
        return {"next": "__end__", "data": state} # Exits the loop
    else:
        state["status"] = "needs_revision"
        return {"next": "writer_agent", "data": state} # Send it back!

# 2. Wire them up
router = MicroAgentRouter()
router.add_node("writer_agent", writer_agent)
router.add_node("reviewer_agent", reviewer_agent)

# 3. Set the entry point and run
router.set_start("writer_agent")
final_state = router.run(initial_data={"topic": "AI Agents"})

print(final_state)
# Output: {'topic': 'AI Agents', 'content': 'This is a great article.', 'status': 'published'}

Features

• Zero Dependencies: Pure standard library.
• Infinite Loop Protection: Built-in max_steps to prevent agents from getting stuck in eternal loops and draining your API budget.
• Transparent State: The state is just a standard Python dictionary passed between your functions. No magic classes.