Creating Agents

Learn how to create effective agents in Converra.

Via Dashboard

1. Go to converra.ai/agents
2. Click New Agent
3. Fill in the details:
   • Name: A descriptive name (e.g., "Customer Support Agent")
   • Model: The LLM you'll use (e.g., gpt-4o, claude-3.5-sonnet)
   • System Prompt: Your agent's system prompt
   • Description: What this agent does (optional)
   • Tags: For organization (optional)

Via MCP (AI Assistant)

Tell your AI assistant:

Create an agent called "Sales Assistant" for gpt-4o with this system prompt:

You are a helpful sales assistant. Your goal is to understand customer
needs and recommend appropriate products. Be friendly but professional.
Always ask clarifying questions before making recommendations.

Via SDK

import { Converra } from 'converra';

const converra = new Converra({
  apiKey: process.env.CONVERRA_API_KEY
});

const agent = await converra.agents.create({
  name: 'Customer Support Agent',
  content: `You are a helpful customer support agent.
Your primary goals are:
1. Resolve issues quickly
2. Be empathetic and understanding
3. Escalate when necessary`,
  llmModel: 'gpt-4o',
  description: 'Main support chatbot',
  tags: ['support', 'production']
});

console.log('Created:', agent.id);

Via API

curl -X POST https://converra.ai/api/v1/agents \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Customer Support Agent",
    "content": "You are a helpful customer support agent...",
    "llmModel": "gpt-4o"
  }'

System Prompt Structure Tips

Be Specific About Role

✓ "You are a customer support agent for an e-commerce company..."
✗ "You are helpful..."

Include Behavioral Guidelines

When handling complaints:
1. Acknowledge the frustration
2. Apologize for the inconvenience
3. Offer a solution
4. Follow up to confirm resolution

Set Boundaries

You should NOT:
- Make promises about refunds without checking policy
- Share internal company information
- Provide legal or medical advice

Add Examples (Few-Shot)

Example interaction:
User: "My order hasn't arrived"
Agent: "I'm sorry to hear that. Let me look into this for you.
Could you please provide your order number?"

Automatic Content Analysis

When you create an agent, Converra automatically runs a content analysis to identify strengths, weaknesses, and improvement opportunities. This also happens when conversations are synced from LangSmith or Langfuse — new agents discovered during sync get analyzed immediately.

You don't need to trigger analysis manually. It runs in the background and results appear on the agent's Insights tab within seconds.

Supported Models

Model	Provider	Notes
gpt-4o	OpenAI	Recommended for most use cases
gpt-4.1	OpenAI	High quality
gpt-3.5-turbo	OpenAI	Fast and cost-effective
gpt-o3	OpenAI	Reasoning model
claude-3.5-sonnet	Anthropic	Excellent reasoning
claude-opus-4	Anthropic	Highest capability
gemini-2.5-pro	Google	Long context

Next Steps

• Managing Agents - Update and organize agents
• Best Practices - Write better system prompts
• Running Optimizations - Improve automatically