What you can check before you hand over a business process.

When a system sends a first message on its own, on chat, email or WhatsApp, the message states that it is an AI. The sending step builds that in. Voice is not built yet. When there is a voice, it will be plainly synthetic, will say it is an AI and will never copy a real person's voice. The answer box on this site searches a fixed list of written answers, so it is not an AI either.

This website's research notes say they were researched and drafted with AI, and name the reviewer if there was one.

Two model families are used, Hermes and DeepSeek, and the code accepts no others. Product by product:

• The AI Outreach & Lead Qualification Engine: the product's own entry says it runs only on Hermes and DeepSeek.
• The E-commerce Support Assistant and the Connected Operations Layer: the stack lists Hermes and DeepSeek.
• Every other product: a person confirms in writing which models are used, because the product entries do not say.

Both families are hosted, and the text they see is redacted as described under data handling, which is not proof that no personal data remains. Whether hosted models may see redacted personal data is not settled, which is why we make no claim that your data never reaches another AI company. Nothing is on file about what the providers' own terms say on training, so we state nothing about it. Nothing here has run live yet.

• Kill switch. All sending passes one checkpoint, so halting the checkpoint halts the sending. Each agent reads the switch before it acts, and if it cannot read it, it does not act.
• Circuit breaker. It stops a failing action from repeating. Only a person resets it.
• STOP. A customer's STOP is honored before anything else is said.
• Caps. A person sets a daily message cap, and it applies to every sending agent.
• Ships switched off. A system starts in a dry run where nothing leaves. Turning it on is one recorded step by a named person.

Not on file yet: who holds the switch for your system day to day, and the cancellation and minimum-term terms. A person confirms both before you sign.

Each action is appended to a log with four things: who acted, what was done, a hash of the content, and the hash of the entry before it. Change or delete an entry and the chain breaks, and the break shows when the chain is checked. The chain has one limit. It proves nothing was altered afterwards, but it does not prove that what the system did was right.

• We do not send cold or broadcast WhatsApp, make cold calls or use bought or scraped lists. Most systems contact only people who wrote to you first or agreed to hear from you. One differs: the AI Outreach & Lead Qualification Engine can write first, only to business contacts from public business data, an introduction or an inbound inquiry, each cleared by a consent check, with AI disclosed on the first touch. Whether cold business email is allowed depends on the country, and the market pages quote each regulator.
• We do not pretend to be a person.
• We do not move money out. Refunds, credit notes, write-offs, payouts, transfers and chargeback replies are not functions the systems have.
• We do not sign contracts or place trades.
• We do not scrape. The search agents read your own site only and obey robots.txt.
• We do not post on social media for you.
• We do not promise rankings, citations, leads, revenue or accuracy.
• We do not decide complaints or legal matters. The system stops after one neutral line and tells you.
• We do not guess. If your data or policy does not cover a question, the system says so and passes it to a person.
• We do not staff a human helpdesk around the clock, and we do not support hardware or IT we did not build.

From the first day, the workflow, the account and the data are yours. Before a system is marked live, we hand you its credentials and documentation in writing, not afterward. That is a gate in our own delivery process rather than a promise alone: a deal cannot be marked live until the handover is logged.

What is not on file: whether every part of every system keeps running without us. A person tells you exactly what keeps running for the product you are looking at. We would rather say that than promise it for all nine.

The country pages, the comparisons, the trends page and the research notes are researched and drafted with AI by Gonzo Research. They are not disguised as human work. The method:

1. Public pages only. We read public pages the way a polite visitor does: robots.txt respected, an honest user agent, low volume, no paywall bypass, and no scraping of search result pages.
2. Every claim has a source and a date. A fact about the outside world links to the page we read and the date we read it. A figure we could not confirm from a primary source is marked in the data, and it is left out of the page rather than published as fact.
3. Short quotes only. At most one short sentence from a source, and we say who said it.
4. Regulator and vendor pages win. Where we could read the regulator or the vendor, we cite them. Where a page blocked us, we did not work around it, and the fact stayed out.
5. Review is shown only if it happened. A note carries a "Reviewed by" line only when a named person has reviewed it.
6. General information, not legal advice. Rules change. Check the linked page and ask a local adviser.

The trends page uses public news and trending feeds and shows nothing it did not get. It shows no search volumes, because real volumes need a licensed source we have not connected.

How it works walks through the seven checks and the autonomy ladder. The doubts page lists the questions a careful buyer asks, with "a person will confirm" wherever we have no answer yet. Compare says where the alternatives win.