Skip to main content
Multi-agent collaboration system with specialized roles.

​
Code

from bindu.penguin.bindufy import bindufy
from orchestrator import Orchestrator
from dotenv import load_dotenv

load_dotenv()

orchestrator = Orchestrator()

def handler(messages: list[dict[str, str]]) -> str:
    if not isinstance(messages, list):
        return "Invalid input format: messages must be a list."
    
    if not messages:
        return "No input message received."
    
    last_msg = messages[-1]
    
    if not isinstance(last_msg, dict):
        return "Invalid message structure."
    
    user_input = last_msg.get("content")
    
    if not user_input or not isinstance(user_input, str):
        return "Empty or invalid message content."
    
    try:
        result = orchestrator.run(user_input)
        return result
    except Exception as e:
        return f"Internal agent error: {str(e)}"

config = {
    "author": "[email protected]",
    "name": "killer-agent-swarm",
    "description": "Multi-agent AI system for deep research, summarization, critique and reflection.",
    "capabilities": {"streaming": True},
    "deployment": {
        "url": "http://localhost:3773",
        "expose": True,
        "cors_origins": ["http://localhost:5173"]
    },
    "skills": ["skills/agent-swarm-intelligence"],
    "storage": {"type": "memory"},
    "scheduler": {"type": "memory"}
}

bindufy(config=config, handler=handler)

#bindufy(config, handler, launch=True)
# This will create a tunnel to your agent and expose it on port 3773

​
How It Works

Agent Roles
  • Planner: Breaks query into structured tasks
  • Researcher: Performs deep factual research
  • Summarizer: Condenses research into clear explanations
  • Critic: Reviews and refines outputs
  • Reflection: Evaluates quality and triggers self-improvement
Orchestrator
  • Coordinates agent execution pipeline
  • Manages task flow: Planner → Researcher → Summarizer → Critic → Reflection
  • Handles retries and error recovery
  • Validates output quality
Execution Flow
  1. Planner decomposes query into tasks
  2. Researcher gathers information
  3. Summarizer creates concise output
  4. Critic reviews and refines
  5. Reflection evaluates quality
  6. If quality is low, retry with improvements
Self-Correction
  • Reflection agent assesses output quality
  • Triggers automatic refinement loops
  • Max 2 retries for quality improvement

​
Run

# Set API key
echo "OPENROUTER_API_KEY=your-key-here" > examples/agent_swarm/.env

# Run agent swarm
uv run examples/agent_swarm/bindu_super_agent.py
Try: “Research about quantum computing and explain its applications” Go to frontend and run npm run dev Open http://localhost:5173 and try to chat with agent swarm