Python JavaScript

Agent Configuration (JavaScript/TypeScript)¶

Learn how to configure and customize AI agents in GraphBit.

Overview¶

GraphBit agents are AI-powered components that can: - Process natural language input - Generate intelligent responses - Execute with custom configurations - Integrate with multiple LLM providers

Creating Agents¶

Basic Agent¶

import { AgentBuilder, LlmConfig } from '@infinitibit_gmbh/graphbit';

const config = LlmConfig.openai({
  apiKey: process.env.OPENAI_API_KEY || '',
  model: 'gpt-4o-mini'
});

const agent = await new AgentBuilder('My Agent', config)
  .build();

const response = await agent.execute('What is quantum computing?');
console.log(response);

Agent with Description¶

const agent = await new AgentBuilder('Research Assistant', config)
  .description('Helps with academic research and analysis')
  .build();

Agent with System Prompt¶

const agent = await new AgentBuilder('Code Reviewer', config)
  .description('Reviews code for best practices')
  .systemPrompt('You are an expert code reviewer. Focus on security, performance, and maintainability.')
  .build();

Agent with Configuration¶

const agent = await new AgentBuilder('Creative Writer', config)
  .description('Generates creative content')
  .systemPrompt('You are a creative writer specializing in short stories.')
  .temperature(0.9)        // Higher for creativity
  .maxTokens(2000)         // Longer responses
  .build();

Configuration Options¶

Temperature¶

Controls randomness in responses (0.0 to 2.0): - 0.0-0.3: Deterministic, factual - 0.4-0.7: Balanced - 0.8-2.0: Creative, varied

.temperature(0.7)

Max Tokens¶

Maximum length of response:

.maxTokens(1000)  // Limit response length

Agent Methods¶

Execute¶

Send a message and get a response:

const response = await agent.execute('Explain machine learning');
console.log(response);

Get Agent Info¶

const name = await agent.name();
const description = await agent.description();

console.log(`Agent: ${name}`);
console.log(`Description: ${description}`);

Complete Examples¶

Research Assistant¶

import { AgentBuilder, LlmConfig } from '@infinitibit_gmbh/graphbit';

async function createResearchAssistant() {
  const config = LlmConfig.openai({
    apiKey: process.env.OPENAI_API_KEY || '',
    model: 'gpt-4o'  // More capable model
  });

  const agent = await new AgentBuilder('Research Assistant', config)
    .description('Analyzes and summarizes research papers')
    .systemPrompt(`You are a research assistant specializing in academic literature analysis. 
      Provide clear, well-structured summaries with key findings and insights.`)
    .temperature(0.3)    // Lower for factual accuracy
    .maxTokens(1500)
    .build();

  return agent;
}

async function main() {
  const agent = await createResearchAssistant();

  const query = 'Summarize the latest developments in quantum computing';
  const response = await agent.execute(query);

  console.log('Research Assistant Response:');
  console.log(response);
}

main().catch(console.error);

Code Analyzer¶

async function createCodeAnalyzer() {
  const config = LlmConfig.openai({
    apiKey: process.env.OPENAI_API_KEY || ''
  });

  const agent = await new AgentBuilder('Code Analyzer', config)
    .description('Analyzes code for improvements')
    .systemPrompt(`You are an expert code reviewer. Analyze code for:
      1. Security vulnerabilities
      2. Performance issues
      3. Best practices
      4. Code smell
      Provide specific, actionable recommendations.`)
    .temperature(0.2)
    .maxTokens(2000)
    .build();

  return agent;
}

Creative Writer¶

async function createCreativeWriter() {
  const config = LlmConfig.anthropic({
    apiKey: process.env.ANTHROPIC_API_KEY || '',
    model: 'claude-3-5-sonnet-20241022'
  });

  const agent = await new AgentBuilder('Creative Writer', config)
    .description('Generates creative stories and content')
    .systemPrompt(`You are a creative writer with expertise in:
      - Short stories
      - Character development
      - Vivid descriptions
      - Engaging narratives`)
    .temperature(0.9)     // High for creativity
    .maxTokens(3000)      // Allow longer stories
    .build();

  return agent;
}

Best Practices¶

Choose Appropriate Models
GPT-4o / Claude Opus: Complex reasoning
GPT-4o-mini / Claude Sonnet: Balanced tasks
GPT-3.5-turbo / Claude Haiku: Simple tasks
Set Clear System Prompts
Define the agent's role and expertise
Specify output format
Include constraints and guidelines
Optimize Temperature
Lower (0.0-0.3) for factual, deterministic tasks
Medium (0.4-0.7) for balanced outputs
Higher (0.8-1.0+) for creative tasks
Manage Token Limits
Set appropriate maxTokens for your use case
Consider model context windows
Balance completeness vs. cost

Error Handling¶

import { AgentBuilder, LlmConfig } from '@infinitibit_gmbh/graphbit';

async function safeAgentExecution() {
  try {
    const config = LlmConfig.openai({
      apiKey: process.env.OPENAI_API_KEY || ''
    });

    const agent = await new AgentBuilder('Assistant', config)
      .build();

    const response = await agent.execute('Hello!');
    console.log(response);

  } catch (error) {
    console.error('Agent execution failed:', error);
  }
}