Privacy Policy
Last updated: August 10, 2026
Rules Lawyer is built to be private by design: your account, your conversations, the rulebook databases, and the files we serve you all live on our own self-hosted infrastructure. The AI models that read and write text — the one that writes your answer, and a few smaller ones around it — run at one outside AI provider, DeepInfra. Everything that goes there is listed below. There is no ad network and no data broker anywhere in this product. We use one product-analytics service (PostHog) to count page views; it never sees your questions, answers, or email.
What we collect
| Data | Why |
|---|---|
| Email, display name | Your account and sign-in |
| Password (stored as an argon2 hash — we never see it) | Authentication |
| Your questions and the answers | Operating the chat; your conversation history |
| Thumbs up/down votes and comments | Improving answer quality |
| Question counts per week | Enforcing fair-use limits |
| Rulebook submissions (URLs) | Growing the game library |
We do not collect payment details — supporter billing happens entirely on Patreon. If you choose to link a Patreon account we store your Patreon user id and email to recognize your membership.
Cookies
One first-party session cookie keeps you signed in. That's the whole list. It is strictly necessary for the Service to function, which is why you don't see a cookie banner. Analytics does not set cookies (it uses browser localStorage with a random identifier, and respects your browser's Do Not Track setting).
Who else sees your data
- Our email provider, to send verification and password-reset emails from bgruleslawyer.com.
- Patreon, only if you link your account, and only the membership status exchange.
- PostHog (product analytics, US cloud): which pages are visited, tied to a random identifier stored in your browser — never the text of your questions or answers, and never your email or name.
- DeepInfra (AI inference, US cloud). Six things go there, and nothing else does:
- Writing your answers. For every account, free and supporter alike, we send your question, the recent messages in that conversation, and the rulebook excerpts we retrieved; their model writes the answer.
- Understanding your question and naming your conversation. Smaller calls that work out which game you're asking about, recognize questions about the library itself, and write the short title in your sidebar. These send your question, and for the title, the answer.
- Automated quality scoring. A sample of roughly one in ten answered exchanges is sent back to be scored for whether the answer addressed the question and cited its sources properly. This is how we notice quality slipping without reading everyone's conversations by hand.
- Searching the rulebooks. Your question is turned into a list of numbers (an "embedding") by their model so we can match it against the rulebook passages in our own database. The search itself, and the passages, stay with us.
- Read-aloud. If you press play on an answer, the text of that answer is sent to be turned into audio. We store the audio on our own hardware and reuse it, so the same answer is never sent twice.
- Processing rulebooks. When a rulebook is added to the library, its text is sent to be checked, read, summarized into the structure that makes it searchable, and turned into the same kind of numbers as above. This is third-party rulebook text, not anyone's conversation: it carries no question, no account, and no personal data.
None of these carries your identity: no email, no name, no account identifier — just the text of the exchange and the rulebook excerpts. DeepInfra's published data policy states that they do not train on API inputs or outputs and do not sell them, and that inputs and outputs may be stored briefly for debugging.
What stays on our hardware: your account, your conversations, the rulebook database and the files behind every citation link — plus two AI models we still run ourselves, the one that scores which rulebook passages are most relevant to your question, and the one that reads figures and diagrams out of rulebook pages when a book is added.
- Nobody else.
If the AI provider is unavailable
We have no second place to write an answer. If DeepInfra is unreachable or failing, we tell you the assistant is offline, and that question is not counted against your weekly limit. If the provider fails partway through an answer, you keep the part that was written, with a note saying it was cut short. We would rather say so plainly than quietly serve you something worse.
Retention and deletion
Your conversations stay until you delete them (deleting a conversation removes its stored messages). You can delete your entire account from the Account page — this permanently removes your profile, conversations, votes, and usage records. Anonymized question/answer records (with your account identifier removed) may be retained to improve answer quality.
Improving the Service
We review flagged answers (for example, ones that received a thumbs-down) and anonymized question/answer logs to fix bad answers and build regression tests. That review is done by us, on our own hardware, and the logs we keep for it live on our own hardware too. The only things that leave it are the flows listed above — none of which carries your identity.
Children
The Service is not directed at children under 13, and we do not knowingly collect their data.
Changes and contact
Material changes will be posted here with a new date. Privacy questions or requests: [email protected].