Introduction

Converra automates AI agent improvement — the fix, test, deploy loop that eats engineering time. Your agents improve while your engineers build.

What is Converra?

Converra generates targeted fixes for your agents, simulation-tests them, and ships what works. Each cycle builds on the last — your agents get measurably better without engineering effort.

Converra helps you:

• Fix agents automatically - Generate targeted prompt, config, and model variants addressing diagnosed issues
• Diagnose failures - Identify why agents underperform, down to the specific step in multi-agent flows
• Prove changes before production - Every fix survives 36+ simulated runs and regression testing
• Ship proven improvements - Deploy automatically via API, webhooks, or GitHub PR. Instant rollback if any metric regresses

How It Works

1. Diagnose - Import runs and traces to identify failure patterns
2. Fix - Generate targeted variants addressing diagnosed issues
3. Prove - Simulate against diverse personas and regression-test against known-good scenarios
4. Deploy - Ship the proven winner back to production — automatically

Key Concepts

Agents

Your AI agents with their system prompts, objectives, constraints, and LLM settings.

Agent Systems

When you import multi-step traces (e.g., a router handing off to specialists), Converra auto-discovers agent systems: agents that operate together in a flow. Converra simulates these systems with a bounded flow model so runs terminate and results stay comparable.

Runs

Real user-AI dialogues logged for performance analysis and insight generation.

Optimizations

Automated improvement cycles that generate prompt variants and evaluate them through simulation.

Insights

Aggregated learnings from runs identifying patterns, issues, and opportunities.

SDK

Wrap your LLM clients (OpenAI, Anthropic) with one line to automatically capture runs and enable A/B testing. Available for Node.js and Python.

Next Steps

• Quick Start - Get up and running in 2 minutes
• Core Concepts - Understand the platform in depth
• SDK Integration - Wrap your LLM clients for automatic capture
• API Reference - Integrate via REST API
• MCP Setup - Use with Claude Code