Privacy and terms

Last updated 30 October 2026. You sign in with a Google or Microsoft account. FS Vector work accounts are approved at once; other new accounts wait until an admin approves them. There are no passwords and no emailed sign-in codes.

What we store

Your account: email address, role, status (including whether it is approved) and the date it was created.
Sign-in methods: the Google or Microsoft account identifier you connected.
Sessions: a hash of your session token, your IP address, your browser's user agent and when you signed in and were last active.
Security log: sign-in events with your account id, IP address and a keyed hash of your email.
Abuse limits: counters keyed by your account id, your IP address or a hash of your email, for request rates, daily AI and search calls, and confirmation mails.
Request log: for each search, the time, your account id (or IP address when you are signed out), the result status, the data version and how long it took. It holds no search text and no request id.
Search questions: the text of each question asked in Search on the site or through the Claude link search tool, when it was asked, your account id (or IP address when you are signed out) and the result status. We use them to find and fix searches that fail. They are kept in a separate file that is never backed up, deleted after 7 days at most, and deleted at once with your account.
Search and Claude link traces: a SHA-256 hash of the question, never its text. Claude link traces also keep the account or token name that made the call.
Claude link: when you make a Claude link we keep a scrambled copy and the last time Claude used it, and delete both when you turn the link off or delete your account.
Alerts: if you subscribe to hearing alerts, your email and which hearings you picked.
Label corrections: if you correct a For, Against or Neutral label in Prepare, we keep your account id (when you are signed out, a keyed hash of your IP address, never the address itself), the quote and position it was on, the label you chose, the words you highlight and when; setting a label back to ours deletes the row. Rows tied to an account are deleted with it; email us to remove signed-out corrections.

We do not keep your name, photo or any token from Google or Microsoft. To answer you, we send text to the AI providers below: your searches and chat questions, and for Prepare, the text of the brief you upload, the client label, the positions, and the purpose, reader, cause, tone and note fields of the memo. Apart from the search questions above, we do not tie that text to your account in storage.

Why

To sign you in and keep your account secure, to stop abuse such as request floods and mail floods, and to send the alerts you ask for.

How long

Sessions: 30 days at most, or 14 days without use.
Security log: 90 days.
Search questions: 7 days at most.
Abuse limits: 24 hours.
Accounts and alert subscriptions: until you delete them.
Sign-up requests no admin has approved: 30 days.

Expired rows are purged every night. We keep a single backup copy of the account database, and each nightly copy replaces the last one, so deleted data leaves the backup within 24 hours. Operational logs and traces are rotated by size, so older entries are overwritten as new ones arrive. Live search traces and logs never store the question text.

Who else sees it

Google and Microsoft, when you sign in with them.
AgentMail, which delivers alert confirmations and alerts to your email.
Anthropic. Its Claude model writes AI summaries, follow-up chat answers and Prepare memos from the text listed above. It does not see who sent the text.
Anthropic, through the Vercel AI Gateway. Its Claude model grades how well each found passage answers a Search question. It sees the question and the public passages, not who sent them.
OpenAI, through the Vercel AI Gateway. In Prepare its model reads the brief, documents, note or cause you give to suggest what to search for. It does not see who sent the text.
OpenAI and Baseten, through the Vercel AI Gateway. In Prepare, Baseten's Mammoth model and OpenAI's model read each public hearing statement found for your positions and say whether the member agrees, disagrees or takes no side. They see the position text, not who sent it.
OpenAI, through the Vercel AI Gateway. It checks the draft memo and its brief context against the selected members and public hearing excerpts for correct person attribution. It also checks each point of an AI summary or follow-up answer against the public hearing statement it cites. It does not see who sent the text.
Voyage, which turns search text into search embeddings, and Qdrant, which stores and searches those embeddings. Neither sees who asked.
Groq, which turns public hearing audio into text. It sees none of your data.
Vercel and our hosting provider, which carry the traffic.

We do not sell data, show ads or use analytics or tracking cookies. The only cookies are the ones that keep you signed in and protect forms.

How to delete your data

Open your account page and choose Delete account. This removes your account, sign-in methods, sessions, alert subscriptions and label corrections straight away; the single backup copy is replaced within 24 hours. To leave alerts only, use the leave link in any alert email. You can also email us to ask for a copy or deletion of your data.

Contact

maxanthonycorbett@gmail.com

Terms of use

Access is personal; do not share your sign-in or let others use your account.

Transcripts, speaker names and summaries are produced automatically from public hearing video and records. They can contain errors and are provided as is, without warranty. Check the official record before you rely on a quote.

Do not try to break its security, overload it, scrape it in bulk or use it to send unwanted mail. We may suspend accounts that do.