Fair use

A connected wallet may ask 50 questions per day. The count resets at midnight UTC, which is the same moment for everybody rather than a rolling window nobody can predict.

The panel tells you what is left once you get near the limit. Below that it stays quiet, because a counter on every question makes a free feature feel metered.

Every answer costs money to produce, paid by this project rather than by you. 50 a day is far more than reading the corpus requires and low enough that the bill stays predictable. It exists to keep the feature free for everyone, not to push anyone towards paying.

If you hit it regularly and it is getting in your way, that is worth knowing: say so on the repository. A limit that the people actually using it keep hitting is the wrong limit.

The limit has to attach to somebody, and the only identity Beersy has is the public key of the BRC-100 wallet you connect. There is no account, no email and no password, so there is nothing to sign up for and nothing to leak.

The wallet is not charged. It is used as a name, not as a payment method. Everything else on the site, including all 162 standards in full, needs no wallet at all.

What you write is never stored. A question you type is sent to the model to be answered and is not written down anywhere here. There is no column for it in the database, which is a stronger guarantee than an intention: storing it later would take a schema change somebody has to make on purpose.

What is recorded, against the public key you connected: that a question was asked, when, which page you were on, whether you typed it or pressed one of the suggested questions, which suggestion if so, and which name for the chain you have selected. Enough to see which parts of the corpus leave people stuck and which suggestions are worth offering.

Also one number, used to enforce the allowance: how many questions that key has asked today. It is replaced by tomorrow’s number when the day rolls over.

Answers come from Anthropic’s API. Their terms cover what happens to a request in transit, and API traffic is not used to train their models.

Each question is sent with the text of the standard you are reading, so answers are grounded in the specification rather than in whatever the model absorbed about Bitcoin generally. That is the difference between a useful answer here and a confident wrong one.

It is still a model, and it can still be wrong. Where an answer and the specification disagree, the specification is right. Every page links to its exact upstream file for that reason.

Automated or bulk use, or working around the limit with generated keys, gets the feature switched off rather than rate limited further. Nothing else about the site changes: reading has never required anything and will not start.