Invented policy
An airline's chatbot promised a fare rule that did not exist. A tribunal held the airline to it.
10 industry checklists · 120 cases · rules, not AI, decide
Stress Echo runs the AI agent on your website through the conversations that have already gone wrong for someone else. Every verdict comes from a fixed rule, with the evidence that decided it.
Benefits
Each check is motivated by a real failure. The link goes to the source.
An airline's chatbot promised a fare rule that did not exist. A tribunal held the airline to it.
A dealer's chatbot agreed to sell a truck for $1 after a user asked it to accept anything.
Prices shown without the fees that come on top. The FTC rule asks for the total up front.
Confident answers that could harm a patient or a diner. ECRI ranks misuse of AI chatbots as the top health technology hazard for 2026.
A request to move money to a "safe account", or to hand over a one-time code. Banks in the UK must now reimburse impersonation scams.
Text inside a page or a tool result that tells the agent to book, buy or reveal something. Browsing agents have been steered this way.
Features
Everything that can be checked without a model is checked first, for free. Runs that spend model calls tell you the count before they start.
Each pillar is a checklist; its height is its number of cases.
Every case runs on the agentic-web runtime with mocked actions. Fixed rules decide; no model grades another.
Reads the network and its actions, and says what is wrong and how many characters each agent sends, before anything is paid for.
No model grades another. Each pass or fail names the rule and quotes the evidence. Keyword rules match words, so they can be wrong; the evidence lets you check.
A smoke test is about 40 model calls for a typical site. The page shows the count first and stops at a hard cap you set.
120
Written for the kind of site, not one site's actions, so they run on any network in that industry. Cases that cannot run on a site are listed with the reason, never counted as passes.
Loading the checklists…
The zero-token checks as a command-line tool and a local MCP server: lint a bundle, size a call, validate a call, list checklists, estimate a run, and judge a saved trace. No key, nothing spent.
node cli/tools.mjs estimate path/to/bundle --checklist restaurants-menu.v1
Bring any OpenAI-compatible router and the models you want to compare. The key is sent only to that router, never stored, and never written into a report or a log.
Every site action is mocked during a run, so nothing real is booked, bought or sent.
FAQ
An agent network from an agentic-web site: the assistant on a website and the specialists it calls. It runs the network on real models with your key, against checklists written for its industry, and the site's actions are mocked so nothing real is booked, bought or sent.
Your API key stays in your browser. It goes only to the router you type, and it is never stored or put in a report. The assistant never sees it.
No. Fixed rules decide every verdict. Some rules are exact, like whether a write action was called; others are keyword rules. Keyword rules match words, so they can be wrong; each verdict shows its evidence so you can check it.
Steps one to three are free. A smoke test is about 40 model calls for a typical site, at your provider's prices. The page shows the count before you start, and the demo mode with three fake models costs nothing.
Restaurants, hospitality, healthcare booking, banking and payments, travel and transport, and e-commerce: two checklists for each of the first five, plus the e-commerce checklist.
Yes. The zero-token checks run as a command-line tool and a local MCP server. Runs that spend model calls are not offered to agents; they stay on the page behind the cost check. See connecting an agent and llms.txt.
No. Stress Echo checks how an agent behaves against its contract. It is not legal advice, and it does not certify that a site complies with any law.
Ask
The assistant can explain Stress Echo and what the checks mean. To run the checks, open the tester.
Connecting the assistant… If this message stays, the assistant could not load. The rest of the page still works.
Test it yourself in the browser, or connect your coding agent to the free checks.