A powerful TypeScript library for building AI agents with tool-use capabilities. Swarm enables you to create intelligent agents that can use multiple tools, chain operations together, and handle complex multi-step tasks.

Inspired by openai/swarm.

• 🛠 Multi-Tool Support: Create agents with access to multiple tools and functions
• 🔄 Tool Chaining: Intelligently chain multiple operations together
• 🌊 Streaming Support: Real-time streaming of agent responses
• 🎯 Type Safety: Full TypeScript support with comprehensive type definitions
• 🔍 Smart Context: Pass and maintain context variables across operations
• ⚡ Parallel Execution: Support for parallel tool calls when possible
• 🐛 Error Handling: Robust error handling and retry mechanisms
• 📝 Logging: Configurable logging system for debugging

You can install the package using npm:

npm install sw4rm

1. Create an agent with tools:

import { Agent } from 'sw4rm/types';

const myAgent: Agent = {
  name: 'MyAssistant',
  model: 'gpt-4o-mini',
  instructions: 'You are a helpful assistant...',
  functions: [
    async function calculate(params: { expression: string }) {
      // Implementation
    }
  ],
  parallelToolCalls: true
};

2. Initialize and run the Swarm system:

import { Swarm } from 'sw4rm/core';

const swarm = new Swarm();
const result = await swarm.run(
  myAgent,
  messages,
  contextVariables,
  { stream: true }
);

3. Use the CLI interface:

import { CLI } from 'sw4rm/cli';

const cli = new CLI({
  agent: myAgent,
  stream: true
});

await cli.start();

Check out the examples directory for complete working examples, including:

• Multi-tool agent implementation
• Weather and web search integration
• Data analysis and unit conversion
• Complex query handling

To run the examples:

npm run start:example

sw4rm/
├── src/
│   ├── core.ts      # Core Swarm implementation
│   ├── cli.ts       # CLI interface
│   ├── types.ts     # TypeScript type definitions
│   ├── errors.ts    # Error handling
│   └── utils.ts     # Utility functions
├── examples/
│   ├── agent.ts     # Example agent configuration
│   ├── functions.ts # Tool implementations
│   ├── index.ts     # Example usage
│   └── README.md    # Example documentation
└── README.md        # This file

The main class for running agents and managing tool execution.

const swarm = new Swarm(config?: SwarmConfig);
await swarm.run(agent, messages, contextVariables, options);

Command-line interface for interactive agent sessions.

const cli = new CLI(config: {
  agent: Agent;
  contextVariables?: Record<string, unknown>;
  stream?: boolean;
  debug?: boolean;
});
await cli.start();

Defines an agent's capabilities and configuration.

interface Agent {
  name: string;
  model: string;
  instructions: string | (() => string);
  functions: Function[];
  parallelToolCalls: boolean;
}

Configuration options for the Swarm system.

interface SwarmConfig {
  logger?: Logger;
  retryConfig?: {
    maxRetries: number;
    initialDelay: number;
    maxDelay: number;
  };
}

The system includes several error types for different scenarios:

• ValidationError: For input validation failures
• APIError: For API communication issues
• ToolExecutionError: For tool execution failures
• TimeoutError: For operation timeouts

1. Tool Implementation

   • Implement proper error handling in tools
   • Return clear, formatted responses
   • Include type definitions for parameters

2. Agent Configuration

   • Provide clear instructions
   • Set appropriate model parameters
   • Consider parallel execution capabilities

3. Error Handling

   • Catch and handle specific error types
   • Provide meaningful error messages
   • Implement retry mechanisms where appropriate

Contributions are welcome! Please feel free to submit a Pull Request.

This project is licensed under the MIT License - see the LICENSE file for details.