# Converra

> Turn agent traces into tested fixes

## Overview

Converra optimizes AI agent prompts through simulation testing outside your production stack. It runs simulated conversations against diverse personas and scenarios, evaluates performance with multi-metric scoring, and recommends winning variants with statistical confidence.

## Core Capabilities

1. **Prompt Analysis** - Identify weaknesses and get improvement recommendations
2. **Simulation Testing** - Run prompts against auto-generated personas and scenarios
3. **A/B Simulation Testing** - Compare two prompts with identical test conditions
4. **Regression Testing** - Validate variants against golden scenarios before deployment
5. **Automated Optimization** - Generate and test variants, select winners based on lift
6. **One-Click Deployment** - Apply winning variants to production prompts

## Integration Methods

- **MCP Server** - Native integration with Claude, Cursor, and AI assistants
- **Node.js SDK** - `npm install converra`
- **REST API** - JSON-RPC 2.0 over HTTPS

## Key Metrics

- Success Score (task completion effectiveness)
- AI Relevancy (response appropriateness)
- User Sentiment (tone and satisfaction indicators)
- Clarity (comprehensibility of responses)
- Safety metrics (hallucination, repetition detection)

## Target Users

- Prompt engineers optimizing AI agent behavior
- AI application developers building conversational products
- ML/Platform teams maintaining production AI systems
- Product managers tracking AI agent performance

## Getting Started

1. Get API key at https://converra.ai
2. Add MCP server or install SDK
3. Create/import prompts
4. Run simulations or trigger optimization
5. Review results and deploy winners

## Links

- Website: https://converra.ai
- Documentation: https://converra.ai/docs
- Login: https://converra.ai/login
- Contact: support@converra.ai

## Full Documentation

For comprehensive API documentation and integration guides, see:
https://converra.ai/llms-full.txt