# QAgent

> Automated AI agent quality testing, hallucination detection, and prompt evaluation platform for developers and AI founders.

QAgent (accessible at https://qagent.in) is a continuous evaluation and quality-assurance platform designed specifically for production AI agents and conversational LLM applications. It automatically runs test suites against AI agent endpoints, scoring responses against user-defined ground truth, business rules, and knowledge bases before problematic responses reach end customers.

## Who It Is For

- **Solo Founders & Engineers**: Building customer-facing AI agents who need confidence that their bot will not hallucinate or leak sensitive policies.
- **AI Agencies & Dev Shops**: Delivering custom AI chatbots to clients who require verifiable test reports and QA audits before release.
- **Product Teams**: Migrating LLM models or system prompts who need regression testing to ensure output quality doesn't degrade.

## Core Capabilities & Evaluation Dimensions

1. **Answer Quality & Correctness**:
   - Evaluates whether the agent accurately answers the user's question without extraneous filler, vagueness, or refusal when answers exist.
   - Evaluated on an absolute 0–100% scale against Expected Behavior.

2. **Hallucination Detection & Ground Truth Fidelity**:
   - Compares agent responses directly against authoritative ground truth documents, FAQs, and business policies.
   - Identifies fabricated claims, invented dates, unsupported numbers, or false policy promises.
   - Specifically accounts for dynamic runtime data (order IDs, tracking numbers, generated coupon codes) to prevent false positives.

3. **Policy Adherence & Safety Guardrails**:
   - Tests compliance with custom business rules (e.g. "Never give financial advice", "Always verify order ID before refunding").
   - Flags prompt injection attempts and jailbreak vulnerabilities.

4. **Escalation & Human Handoff Testing**:
   - Verifies whether the agent recognizes frustrated, angry, or high-risk users and triggers appropriate human handoff or support escalation procedures.

5. **RAG (Retrieval-Augmented Generation) Metrics**:
   - **Context Faithfulness**: Ensures the generated answer relies strictly on retrieved context chunks rather than unverified LLM pre-training knowledge.
   - **Contextual Relevancy**: Quantifies how relevant retrieved document chunks are to the user prompt, awarding top scores when rank #1 chunks answer the query.
   - **Context Recall**: Verifies whether all critical facts from ground truth were retrieved and utilized.

6. **Multi-turn Conversation Memory Testing**:
   - Executes multi-turn dialog simulations to verify that the agent remembers user preferences, constraints, and credentials established in earlier turns.

## Pricing Tiers

- **Free Tier**: ₹0/month. Includes 1 connected AI agent, 100 monthly test evaluations, single-turn & multi-turn testing, and full evaluation analytics.
- **Solo Tier**: ₹2,499/month. Includes 5 connected AI agents, 1,500 monthly test evaluations, priority background queue, RAG metrics, and dedicated support.

## Key Links & Resources

- [Homepage](https://qagent.in): Overview of QAgent's continuous agent testing platform.
- [How It Works](https://qagent.in/#how-it-works): Deep-dive into evaluation methodologies and scoring rubrics.
- [Pricing](https://qagent.in/#pricing): Transparent subscription tiers and quotas.
- [Web Application / Sign In](https://qagent.in/login): Sign in via Google OAuth or Magic Link.
- [Terms of Service](https://qagent.in/terms): Usage policies, quotas, and service commitments.
- [Privacy Policy](https://qagent.in/privacy): Zero-data-retention policy on sensitive customer prompts.
- [Cancellation & Refund Policy](https://qagent.in/refund-policy): 7-day refund guarantee details.
- [Contact Support](https://qagent.in/contact): Support inquiries and feedback (qagenttesting@gmail.com).
- [Full Technical Reference](https://qagent.in/llms-full.txt): Complete documentation, scoring rubrics, and platform architecture for LLMs.