# ClientCoded

> ClientCoded is a QA platform for AI agents. We generate adversarial test scenarios, synthetic data environments, and prompt-fix suggestions, then score every conversation and flag exactly what broke, before your users hit it.

ClientCoded tests two kinds of AI agents. Conversational agents (support bots, AI SDRs, lead qualifiers, email agents, onboarding bots) are tested with adversarial synthetic users running real multi-turn conversations, scored pass or fail across 10 dimensions, with regressions flagged after every update. Data and knowledge agents (CRM copilots, ticketing assistants, knowledge-base bots, internal helpdesks) are tested against pre-built synthetic environments with computed ground truth, so answers are graded against what is actually correct rather than what sounds right. ClientCoded also monitors production conversations in real time and alerts when quality drops. We never touch your production data: for data agents, your agent queries our synthetic environments, and we only need a schema, not your records. We do not train models on your data.

## Documentation

- [Integration docs](https://clientcoded.com/docs): How to connect your agent. The production monitoring webhook (POST your conversation transcript, no SDK required), the optional Python SDK for internal tracing, the Query API for data agent testing against synthetic environments, and how to get your agent_id and team api_key.

## Products

- [Adversarial Testing](https://clientcoded.com/agentproof): Generate adversarial personas that push your agent off-script, score every conversation pass or fail across 10 dimensions, and detect regressions after every prompt change.
- [Production Monitoring](https://clientcoded.com/monitoring): Score every production conversation in real time with rule-based checks and AI scoring, and get Slack alerts when quality drops. One webhook, no SDK required.
- [Test Environments](https://clientcoded.com/environments): 35+ pre-built synthetic data environments (Salesforce, Jira, Stripe, Zendesk, GitHub, HubSpot, and more), each with a synthetic dataset, 200 adversarial queries, and computed ground truth for every answer.
- [Pricing](https://clientcoded.com/pricing): Starter at $49/month (monitoring plus data agent testing), Team at $599/month, and custom Enterprise. The first adversarial test is free.

## Framework and methodology

- [The 10-Dimension Framework](https://clientcoded.com/framework): How ClientCoded scores AI agents, across 10 evaluation dimensions, 42 persona archetypes, and 7 industries, with an MCP install to reference the framework inside your IDE.
- [How testing works, and where it breaks](https://clientcoded.com/how-testing-works): The methods teams use to test AI agents today (ground truth comparison, LLM-as-judge, regression tracking, dedicated QA teams), the gaps every method shares, and how adversarial testing and synthetic environments with computed ground truth differ.

## Blog

- [How to QA a Data Agent Without Building Test Datasets by Hand](https://clientcoded.com/blog/how-to-qa-data-agents): Testing agents that query your data by generating synthetic datasets with known-correct answers instead of hand-building test sets.
- [How to Test Your Rasa Agent Beyond the Happy Path](https://clientcoded.com/blog/how-to-test-your-rasa-agent): Testing a Rasa agent with adversarial scenarios across 7 categories, scored on 10 dimensions, with regression checks after every retrain.
- [How to Test Your OpenClaw Agent Before Production](https://clientcoded.com/blog/test-your-openclaw-agent): How adversarial testing catches multi-turn, fabrication, scope, and unauthorized-action failures before users do.
- [Three Layers of AI Agent Evaluation](https://clientcoded.com/blog/three-layers-of-evaluation): Output scoring, production monitoring, and adversarial testing, and why you need all three.
- [We Tested 150+ AI Agents. Here's Where They Break.](https://clientcoded.com/blog/we-tested-150-ai-agents): The average agent fails half its adversarial conversations and breaks between turn three and turn five.
- [What Happens When AI Agents Face Real Users](https://clientcoded.com/blog/what-happens-when-agents-face-real-users): The failures that show up when synthetic adversarial users behave like real customers.
- [Why Ground Truth Testing Fails for AI Agents](https://clientcoded.com/blog/why-ground-truth-testing-fails): Where ground-truth comparison breaks down for conversational agents, and what works better.
- [Why Tracing Alone Doesn't Catch AI Agent Failures](https://clientcoded.com/blog/why-tracing-doesnt-catch-agent-failures): Tracing shows what happened, not whether the outcome was right; behavioral testing plus tracing closes the loop.

## Optional

- [Blog index](https://clientcoded.com/blog): Field notes on AI agent quality.
- [Terms of Service](https://clientcoded.com/terms)
- [Privacy Policy](https://clientcoded.com/privacy)
- Contact: travis@clientcoded.com