# Agent Torture Lab > AI agent crash testing for builders, small teams, and agencies. Agent Torture Lab runs messy customer simulations against AI agents and turns failures into a launch report with transcript evidence, severity, confidence, recommended fixes, and a retest plan. - What it is: a pre-launch crash test for customer-facing text chatbots and AI agents. The output is a launch report with transcript evidence, severity, recommended fixes, and a retest plan. - Inputs: a public API endpoint, a website chat widget (compatibility-gated), or a pasted transcript. - Pricing: the preview is free. The one-time full report is $29 (launch price). One-time unlock includes one free retest: re-run the same target once within your 14-day report window. Automated API and website roasts only — a pasted transcript has no live target to re-run. Agencies buy prepaid credits. - Scoring: deterministic rules produce the score. A Bot Roast sends up to 12 authored scenarios. Any optional AI-assisted findings are listed separately and never change the score. - Built for: self-serve builders, small teams, and agencies shipping AI agents for support, sales, ecommerce, or checkout. - Not: a voice-agent tester, production monitoring, a subscription, a penetration test, or a complete security assessment. Site: https://www.agenttorture.com/ Contact: hello@agenttorture.com Last updated: 2026-10-14 Public marketing pages may be crawled and summarized. Authenticated workspace pages, project setup pages, audit internals, auth callback routes, and onboarding flows should not be treated as public source material. ## Product - [AI agent testing and Bot Roast reports](https://www.agenttorture.com/): Crash test customer-facing AI agents with adversarial chatbot scenarios, transcript evidence, and launch-readiness reports. - [Bot Roast](https://www.agenttorture.com/bot-roast): Set up a crash test for an AI agent and turn messy scenarios into a launch report. - [Agent Torture Lab pricing](https://www.agenttorture.com/pricing): Agent Torture Lab pricing: the anonymous Bot Roast is free. Get the full report with a one-time payment. - [Agency AI agent testing](https://www.agenttorture.com/agencies): Crash test client AI agents and prepare a client-ready report before handoff. - [Sample AI agent launch report](https://www.agenttorture.com/sample-report): Review a sample AI agent crash test report with findings, transcript evidence, and fixes. - [Sample API Agent Roast report](https://www.agenttorture.com/sample-api-report): Review a sample of the live API endpoint Bot Roast report with findings, transcript evidence, and fixes. - [How Agent Torture Lab tests AI agents](https://www.agenttorture.com/how-we-do-it): How adversarial customer simulations become transcript-backed findings, launch scores, fixes, and retest plans. - [About Agent Torture Lab: AI agent crash testing](https://www.agenttorture.com/about): Why Agent Torture Lab exists: catch the AI agent failures that cost trust before real customers do. - [AI agent launch report](https://www.agenttorture.com/reports/ai-agent-launch-report): What an AI agent launch report should include: transcript evidence, launch recommendation, severity, fixes, and retest guidance. ## Guides - [Chatbot test scenarios](https://www.agenttorture.com/resources/chatbot-test-scenarios): Reusable chatbot test scenarios for stress-testing support, sales, ecommerce, and service AI agents. - [Chatbot QA checklist](https://www.agenttorture.com/resources/chatbot-qa-checklist): Run these 20 tests before launch. A pre-launch checklist for chatbot testing before deployment: policy, privacy, injection, memory, tools, and conversion. - [Chatbot security testing checklist](https://www.agenttorture.com/resources/chatbot-security-testing-checklist): Chatbot security testing checklist: 12 checks tagged to the OWASP LLM Top 10, a severity table, vulnerability assessment vs pentest, and retail chatbot risks. - [AI chatbot fails](https://www.agenttorture.com/resources/funny-ai-agent-fails): 11 dated, sourced AI chatbot fails, from Air Canada and the Chevy $1 Tahoe to DPD, each mapped to the pre-launch test that catches it. Free CSV and JSON. - [Sales chatbot test cases](https://www.agenttorture.com/resources/sales-chatbot-test-cases): Sales chatbot test cases for lead qualification, pricing, handoff, buyer objections, conversion dead ends, and hallucinated offers. - [Prompt injection and red-team testing for chatbots](https://www.agenttorture.com/resources/prompt-injection-testing-for-chatbots): Prompt injection and red-team testing for customer-facing chatbots: OWASP LLM01 and LLM06 test classes, safe reporting, and what the free preview covers. - [How to test a chatbot](https://www.agenttorture.com/resources/how-to-test-my-chatbot): How to test a chatbot in 8 steps: scope the risks, write positive, negative and edge cases, script multi-turn and pressure tests, log evidence, and retest. - [AI chatbot audit](https://www.agenttorture.com/resources/ai-chatbot-audit): What an AI chatbot audit covers: policy, privacy, safety, prompt-injection resistance, escalation, conversion, and transcript-backed fixes. - [Are chatbot promises legally binding?](https://www.agenttorture.com/resources/are-chatbot-promises-legally-binding): Moffatt v. Air Canada, the Chevy $1 Tahoe and NYC's MyCity bot: what each case showed about chatbot promises, a 10-question self-check, and what to test. - [Questions to test a chatbot](https://www.agenttorture.com/resources/questions-to-test-a-chatbot): 50 copy-paste questions to test a chatbot across refunds, pricing, privacy, handoff and more, each with the safe behaviour to expect. Free CSV. ## Use cases - [AI customer service agent evaluation](https://www.agenttorture.com/use-cases/ai-customer-service-agent-evaluation): Evaluate customer service AI agents for accuracy, escalation, policy adherence, privacy, tone, and real support outcomes before launch. - [Ecommerce AI agent testing](https://www.agenttorture.com/use-cases/ecommerce-ai-agent-testing): Crash test ecommerce AI agents for refund abuse, discount pressure, checkout confusion, hallucinated policies, and unsafe product claims. - [AI chatbot QA testing](https://www.agenttorture.com/use-cases/ai-chatbot-qa-testing): Test your chatbot API with a free QA report preview. Get transcript evidence, severity-ranked failures, and fixes. No account needed to start. ## Comparisons - [Evals vs monitoring vs pre-launch QA: which do you need?](https://www.agenttorture.com/compare/generic-llm-evals): Evals test a model against a spec, monitoring watches live traffic, pre-launch QA tests a bot before customers do. Compare all three and pick what you need. - [AI chatbot testing tools for customer-facing agents](https://www.agenttorture.com/compare/ai-chatbot-testing-tools): Compare 9 chatbot testing tools, from Promptfoo, DeepEval and Garak to Botium, Cekura, Hamming and Coval, by pricing model, licence, free tier and channels. - [Cekura alternative for one-time chatbot launch reports](https://www.agenttorture.com/compare/cekura-alternative): Compare Agent Torture Lab with Cekura for testing customer-facing chatbots: setup, report-first output, one-time pricing, and who each tool fits. - [Botium alternative for no-setup chatbot testing](https://www.agenttorture.com/compare/botium-alternative): Compare Agent Torture Lab with Botium (Cyara) for chatbot testing: open-source Botium Core, Cyara's platform, pricing, and a report-first launch test. - [TestMu AI alternative for a one-off chatbot launch report](https://www.agenttorture.com/compare/testmu-alternative): Compare Agent Torture Lab with TestMu AI for testing customer-facing chatbots: scenario breadth, credit pricing, no-code setup, and a one-time launch report. ## Methodology - [AI agent testing methodology](https://www.agenttorture.com/methodology): How Agent Torture Lab designs adversarial customer simulations, captures evidence, and turns failures into launch-ready fixes. - [AI agent launch report scoring](https://www.agenttorture.com/methodology/scoring): How Agent Torture Lab scores AI agent failures by severity, confidence, evidence quality, and launch risk. - [AI agent testing glossary: QA, evals and reports](https://www.agenttorture.com/glossary): Plain-English definitions for AI agent testing, scenario packs, transcript evidence, prompt injection, launch reports, and retesting. ## Machine-readable - [Pricing markdown](https://www.agenttorture.com/pricing.md): Parseable pricing, plan status, report unlock, retention, and account requirements. - [Capabilities markdown](https://www.agenttorture.com/capabilities.md): Parseable list of live, gated, unsupported, and planned capabilities. - [Methodology markdown](https://www.agenttorture.com/methodology.md): Parseable testing methodology, evidence standard, scenario families, and limitations. - [AI chatbot fails dataset (JSON)](https://www.agenttorture.com/resources/chatbot-fails-dataset.json): Dated, sourced public chatbot incidents with failure class, consequence, and the test that catches each. CC BY 4.0 — Agent Torture Lab — n=11 public incidents, not a representative sample. - [AI chatbot fails dataset (CSV)](https://www.agenttorture.com/resources/chatbot-fails-dataset.csv): The same incidents as CSV. - [Questions to test a chatbot (CSV)](https://www.agenttorture.com/resources/questions-to-test-a-chatbot.csv): Single test questions grouped by risk family, each with the safe behaviour to expect. CC BY 4.0 — Agent Torture Lab. ## Optional - [AI agent testing resources, guides and checklists](https://www.agenttorture.com/resources): Guides, checklists, scenario categories, glossary terms, and use-case pages for testing AI agents before launch. - [AI agent testing use cases for customer-facing bots](https://www.agenttorture.com/use-cases): Use-case guides for testing support AI agents, ecommerce agents, sales chatbots, and client AI agents before launch. - [AI agent testing tools and chatbot QA comparisons](https://www.agenttorture.com/compare): Compare AI agent testing, chatbot QA, manual review, generic LLM eval tools, and red-teaming options for customer-facing bots. - [Privacy Policy](https://www.agenttorture.com/privacy): What Agent Torture Lab collects, how we use it, how long we keep it, and how to request deletion. - [Terms of Service](https://www.agenttorture.com/terms): Terms for using Agent Torture Lab: service description, authorisation to test, payment, and liability.