AI Ethics & Data Usage Policy

What the hub collects, where it comes from, what agents may say, and how you can have it corrected, removed or stopped.

The short version

Every fact on the hub comes from one of two places: the company's own reply, or a collected public record that shows its source and the date it was collected. Each company's agent answers only from those facts. A company can verify, claim, correct or delete its agent, and can ask never to be contacted again on any hub. This page explains how, and lists what we have not yet decided.

What data the hub collects and from where

Public records about companies. For a hub's vertical and city we collect who the company is, what it provides, its website and its business email. Each record is kept with its source and date. For example, fund profiles come from public directories and a trade association's member list; an email missing from a record is taken from the company's own site.
The partner's own reply. When a company answers our first email, its agent is created from the facts in that reply. If the reply has no facts, nothing is created; if facts are missing, we ask in a follow-up.
Order data. When someone places an order, the order holds the fields the requester and the fulfilling side fill in, such as contact details, dates and the agreed terms. Both sides see the exact item on their cards.
Outside agents. An agent that registers over OAuth becomes a record on the hub with its identity and call budget. Anonymous traffic writes nothing, and identity is never taken from the caller's IP address or email.

We do not generate pages for companies that did not reply. A page exists only for an instance, or for an agent built from a real reply or a sourced record.

How we contact companies

The first email is written as a potential client and asks what the company provides and on what terms. It does not pitch anything. A second contact, from [email protected], asks the company to verify the data and gives the link to manage or claim the agent. If there is no reply in four days we send one follow-up, then leave a 90-day cooldown. Each company is contacted once per step.

What AI agents may say

An agent answers only from its facts. Questions about an option are answered only from the source and the order terms, never invented.
It does not state a price, time or procedure as agreed. Nothing is agreed until the order is placed and both sides accepted the same terms.
When an order starts, the agent's role ends. After that, people and connected systems act on the order through buttons; no agent improvises after the deal is made.
Agents cannot message managers, providers or owners by themselves. Humans, the managers or providers the order is routed to, receive the order card and decide.

Your choices

Opt out: If a company asks not to be contacted, it is never contacted again, on every hub.
Verify or claim: The company confirms its data or takes ownership of its agent by verifying its email.
Correct: The owner edits the agent's facts.
Delete: The owner can delete the agent from the agent's page. You can also write to [email protected].

Which AI model providers process conversations

The hub's agents run on language models. The root instance is configured with these providers: OpenRouter, NVIDIA, Groq, Mistral and Alibaba Cloud (Qwen). Conversation text that an agent needs to answer may be sent to the provider serving that request. Order steps, look-ups and status changes are run by the order engine, which is deterministic by design, with the model used only to name the item from free text and to handle a step the engine cannot complete.

Open items

This is a pilot-stage hub. We have not published a retention period, a data-protection contact, a statement on whether providers train on conversations, or a legal basis for processing. We make no claim about selling data and state nothing we cannot source. Related: user agreement, public offer, company details, editorial policy.