agentic-swarm

Build immortal, self-healing multi-agent AI systems

Features • Installation • Quick Start • Documentation • Contributing

---

What is Agentic Swarm?

Agentic Swarm is a production-grade SDK for building multi-agent AI systems where agents:

• Never die — Auto-heal on failure with state recovery
• Spawn children — Dynamically create sub-agents for complex subtasks
• Route intelligently — Pick the optimal LLM based on task complexity
• Remember everything — Tiered memory with vector-indexed long-term storage
• Stay secure — Audit logs, encryption, RBAC, and data isolation

Highlights

Features

Installation

pip install agentic-swarm

With optional providers:

pip install agentic-swarm[openai]      # OpenAI
pip install agentic-swarm[anthropic]   # Anthropic
pip install agentic-swarm[bedrock]     # AWS Bedrock
pip install agentic-swarm[all]         # Everything

Quick Start

import asyncio
from agentic_swarm import Agent, Swarm, tool

@tool
def search_web(query: str) -> str:
    """Search the web for information."""
    return f"Results for: {query}"

agent = Agent(
    name="researcher",
    role="Research and gather information",
    tools=[search_web],
)

swarm = Swarm(agents=[agent])

async def main():
    result = await swarm.run("Find information about AI agents")
    print(result)

asyncio.run(main())

With LLM Provider

from agentic_swarm import Agent
from agentic_swarm.llm import LLMRouter, OpenAIProvider

router = LLMRouter(strategy="quality_optimized")
router.register_provider("openai", OpenAIProvider(model="gpt-4o"))

agent = Agent(
    name="assistant",
    role="Helpful AI assistant",
    llm_router=router,
)

result = await agent.run("Explain quantum computing")

Never-Forget Memory

from agentic_swarm import Agent
from agentic_swarm.vectordb import InMemoryVectorDB
from agentic_swarm.rag.embedder import Embedder

# Agent that remembers across sessions
agent = Agent(
    name="assistant",
    role="Helpful assistant",
    vectordb=InMemoryVectorDB(),
    embedder=Embedder(),
    auto_archive=True,
    auto_inject_memories=True,
)

# Store memories
await agent.remember("User prefers dark mode")

# Search memories
memories = await agent.recall("user preferences")

# Run task - automatically injects relevant memories
result = await agent.run("What are my preferences?")

Documentation

Examples

# Full SDK showcase
python examples/full_showcase.py --provider openai

# With AWS Bedrock
python examples/full_showcase.py --provider bedrock

# Demo mode (no LLM)
python examples/full_showcase.py --skip-llm

See examples/ for working examples covering all features.

Contributing

We welcome contributions! Please see our contributing guidelines.

Development Setup

git clone https://github.com/nik0811/agentic-swarm.git
cd agentic-swarm
python -m venv env && source env/bin/activate
pip install -e ".[dev,all]"
pytest tests/ -v

Roadmap

• 7 LLM Providers (OpenAI, Anthropic, Bedrock, Gemini, Groq, Ollama, vLLM)
• Tiered Memory (Core + Recall + Archival)
• RAG Pipeline with Hybrid Search
• Agent Sandbox & Data Isolation
• Never-Forget Memory
• Web UI Dashboard
• Distributed Multi-Node Swarms
• OpenTelemetry Integration

License

Apache License 2.0 — see LICENSE for details.

---

_{Built with ❤️ for the AI community}