Creating Agents

The SmythOS SDK offers flexible ways to create and configure agents. This guide covers everything from basic agent creation to advanced model configuration.

The example scripts in examples/01-agent-code-skill, examples/02-agent-smyth-file, and examples/03-agent-workflow-components provide hands-on illustrations of all the concepts covered here.

Code-Based Agent Creation

You can define agents programmatically by instantiating the Agent class with configuration options:

import { Agent } from '@smythos/sdk';

const agent = new Agent({
    id: 'crypto-assistant', // Optional: unique identifier
    name: 'CryptoMarket Assistant',
    behavior: 'You are a crypto price tracker...',
    model: 'gpt-4o', // The language model to use
});
Copy

Configuration Options:

• id (optional): A unique identifier for the agent. Useful for persistence and tracking.
• name: A descriptive name for the agent.
• behavior: Instructions that define the agent's persona and role.
• model: The language model to use (see Model Configuration below).
• mode (optional): Agent execution mode. See Agent Modes.

Importing from .smyth Files

For complex agents with visual workflows created in the SmythOS Builder, you can import pre-configured .smyth files:

import { Agent, Model } from '@smythos/sdk';
import path from 'path';

const agentPath = path.resolve(__dirname, './my-agent.smyth');

const agent = Agent.import(agentPath, {
    model: 'gpt-4o', // Override the model
    teamId: 'team-123', // Optional: specify team context
});
Copy

The .smyth file format allows you to define complex workflows with multiple components, skills, and integrations visually, then import them into your code with full programmatic control.

Model Configuration

SmythOS provides multiple ways to configure language models, from simple to advanced:

Simple Model Configuration

The easiest way is to specify a model by name. The SDK keeps an up-to-date list of popular models:

const agent = new Agent({
    name: 'My Agent',
    behavior: 'You are a helpful assistant',
    model: 'gpt-4o', // Simple string notation
});
Copy

Supported model names include:

• OpenAI: 'gpt-4o', 'gpt-4-turbo', 'gpt-3.5-turbo', etc.
• Anthropic: 'claude-4-sonnet', 'claude-3.5-sonnet', 'claude-3-opus', etc.
• Google: 'gemini-pro', 'gemini-1.5-pro', etc.
• And many more...

Provider-Specific Configuration

For more control over model parameters, use the Model factory with provider-specific methods:

Simple Provider Notation

import { Model } from '@smythos/sdk';

const agent = new Agent({
    name: 'My Agent',
    behavior: 'You are a helpful assistant',
    model: Model.OpenAI('gpt-4o'),
});
Copy

Advanced Provider Notation

Pass custom parameters for fine-tuned control:

import { Model } from '@smythos/sdk';

const agent = new Agent({
    name: 'My Agent',
    behavior: 'You are a helpful assistant',
    model: Model.OpenAI('gpt-4o', {
        temperature: 0.7, // Control randomness (0-2)
        topP: 0.9, // Nucleus sampling
        inputTokens: 200000, // context window size
        outputTokens: 8096, // Maximum tokens the model can generate
        maxTokens: 2000, // Maximum allowed output tokens
        frequencyPenalty: 0.0, // Reduce repetition of token sequences (0.0 - 2.0)
        maxThinkingTokens: 1024, // Maximum tokens to think (reasoning models only)
        presencePenalty: 0.0, // Encourages talking about new topics (0.0 - 2.0)
        stopSequences: ['\n\n'], // Stop sequences
        baseURL: 'https://api.openai.com/v1', // Base URL (for custom endpoints)
        topK: 0, // Top K sampling
    }),
});
Copy

Available Providers:

• Model.OpenAI(...) - OpenAI models (GPT-4, GPT-3.5, etc.)
• Model.Anthropic(...) - Claude models
• Model.GoogleAI(...) - Gemini models
• Model.Groq(...) - Groq inference engine
• Model.DeepSeek(...) - DeepSeek models
• Model.TogetherAI(...) - TogetherAI models
• Model.Ollama(...) - Local models via Ollama
• Model.xAI(...) - xAI models (Grok)
• Model.Perplexity(...) - Perplexity models

For a complete guide on model configuration, available parameters, and best practices, see the dedicated Models Documentation.

Next Steps

Now that you know how to create agents, learn how to give them capabilities with Skills.