SwarmX (forked from OpenAI's Swarm)

An extreme simple framework exploring ergonomic, lightweight multi-agent orchestration.

Highlights

1. SwarmX is both Agent and Workflow
2. MCP servers support
3. OpenAI-compatible streaming-server
4. Workflow import/export in JSON format

Star History

Quick start

SwarmX automatically loads environment variables from a .env file if present. You can either:

1. Use a .env file (recommended):

   # Create a .env file in your project directory
   echo "OPENAI_API_KEY=your-api-key" > .env
   echo "OPENAI_BASE_URL=http://localhost:11434/v1" >> .env  # optional
   uvx swarmx  # Start interactive REPL
   

2. Set environment variables manually:

   export OPENAI_API_KEY="your-api-key"
   # export OPENAI_BASE_URL="http://localhost:11434/v1"  # optional
   uvx swarmx  # Start interactive REPL
   

API Server

You can also start SwarmX as an OpenAI-compatible API server:

uvx swarmx serve --host 0.0.0.0 --port 8000

This provides OpenAI-compatible endpoints:

• POST /chat/completions - Chat completions with streaming support
• GET /models - List available models

Use it with any OpenAI-compatible client:

import openai

client = openai.OpenAI(
    base_url="http://localhost:8000",
    api_key="dummy"  # SwarmX doesn't require authentication
)

response = client.chat.completions.create(
    model="agent-created-by-yourself",
    messages=[{"role": "user", "content": "Hello!"}]
)

Installation

Requires Python 3.11+

$ pip install swarmx # or `uv tool install swarmx`

Usage

import asyncio
from swarmx import Swarm, Agent

client = Swarm()

def transfer_to_agent_b():
    return agent_b


agent_a = Agent(
    name="Agent A",
    instructions="You are a helpful agent.",
    functions=[transfer_to_agent_b],
)

agent_b = Agent(
    name="Agent B",
    model="deepseek-r1:7b",
    instructions="你只能说中文。",  # You can only speak Chinese.
)


async def main():
    response = await client.run(
        agent=agent_a,
        messages=[{"role": "user", "content": "I want to talk to agent B."}],
    )

    print(response.messages[-1]["content"])


asyncio.run(main())

Context Variables

SwarmX supports special context variables that control agent behavior:

background

Provides additional context for the agent. This can be used for:

• Adding external knowledge (web search results, database queries)
• Compressing previous conversation history into summaries
• Isolating context for sub-agents who don't need full conversation history

Example:

context = {
    "background": "Recent news: AI conference announced for next month. User is interested in AI developments."
}

message_slice

Controls which messages are sent to the LLM using Python slice syntax. This enables:

• Context compression by sending only recent messages
• LLM-driven filtering decisions
• Memory management for long conversations

Slice patterns:

• ":10" - First 10 messages
• "-5:" - Last 5 messages
• ":0" - No messages (useful with background for context compression)
• "2:8" - Messages from index 2 to 7

Example:

context = {
    "message_slice": "-10:"  # Send only last 10 messages
}

tools

Dynamically selects which tools are available for the current completion. This allows:

• Context-aware tool selection
• Reducing tool overload by showing only relevant tools
• Dynamic tool routing based on conversation context

Example:

context = {
    "tools": [
        {
            "type": "function",
            "function": {
                "name": "search_web",
                "description": "Search the web for information",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "query": {"type": "string"}
                    },
                    "required": ["query"]
                }
            }
        },
        {
            "type": "function", 
            "function": {
                "name": "get_weather",
                "description": "Get weather information for a location",
                "parameters": {
                    "type": "object",
                    "properties": {
                        "location": {"type": "string"}
                    },
                    "required": ["location"]
                }
            }
        }
    ]
}

Advanced Usage Examples

Context Compression:

# Compress history into background and send no previous messages
context = {
    "background": "Previous conversation summary: User asked about weather, then travel plans.",
    "message_slice": ":0"  # No previous messages
}

RAG Pattern:

# Add web search results to background, send all messages for comprehensive context
context = {
    "background": "Web search results: Latest AI developments from yesterday's conference...",
    # No slice means all message might pass to LLM
}

Dynamic Tool Selection:

# Based on conversation topic, show only relevant tools
if "weather" in user_message:
    context = {"tools": [{"type": "function", "function": {"name": "get_weather", "description": "Get weather", "parameters": {"type": "object", "properties": {"location": {"type": "string"}}, "required": ["location"]}}}]}
elif "search" in user_message:
    context = {"tools": [{"type": "function", "function": {"name": "search_web", "description": "Search web", "parameters": {"type": "object", "properties": {"query": {"type": "string"}}, "required": ["query"]}}}]}

Architecture

graph TD
   classDef QA fill:#ffffff;
   classDef agent fill:#ffd8ac;
   classDef tool fill:#d3ecee;
   classDef result fill:#b4f2be;
   func1("transfer_to_weather_assistant()"):::tool
   Weather["Weather Assistant"]:::agent
   func2("get_weather('New York')"):::tool
   temp(64):::result
   A["It's 64 degrees in New York."]:::QA
   Q["What's the weather in ny?"]:::QA --> 
   Triage["Triage Agent"]:::agent --> Weather --> A
   Triage --> func1 --> Weather
   Weather --> func2 --> temp --> A