Demo
What this demo covers
This is a plain-English lookup over a bounded synthetic dataset of messages and signals. Type a description, or start from one of the examples. A model reads your words only to find which approved entity they refer to.
The model cannot fetch facts, cannot choose for you, and does not answer questions outside this dataset. When it finds candidates, you pick one; the facts shown under “From the database” come only from that click.
When a lookup reaches a boundary — no approved match, a scope that is not covered, or one it cannot settle — the page says which, instead of guessing. Those outcomes are the design working, not the system failing.
How to try it
- Start from an example, or type a description in English, and send it.
- If the page lists candidate scopes (versions of the data), choose one with “Use this scope”.
- On the match you mean, press its “Fetch …” button, such as “Fetch the message's facts”. The facts appear under “From the database”, with their evidence.
Before you send a message
What you type here is sent to Anthropic, the model provider, so its model can find which synthetic entity your words refer to. The relay that forwards it stores no message text and logs none.
This site keeps the conversation in this page's memory only and saves none of it; reloading the page clears it.
To limit how many messages can be sent from one IP address, the relay uses your address; the relay's own request log records neither your address nor your messages. Microsoft Azure, which hosts the relay and the database service this page fetches facts from, may process and log connection details such as your IP address to run the service.