Privacy

Dray reads your Gmail so Claude can search it and draft replies for you. It does not keep your messages. What it keeps is a set of measurements about how you write, the rules it learns from the edits you make, and the identifiers it needs to fetch a message again later.

Dray is operated by Saranyx LLC. It is not affiliated with, endorsed by, or sponsored by Google.

When you connect a mailbox, you grant Dray the Google scope gmail.modify, which allows reading, searching, drafting, sending, labelling and moving mail to Trash. Dray never requests https://mail.google.com/, the scope that would permit permanent deletion, and has no code path that could permanently delete anything.

Dray also receives your Google account identifier, email address and name, which is how it recognises you when you return.

• ›Your Google refresh token, encrypted with AES-GCM before it is written to storage.
• ›A style card per relationship type: distributions of how you open and close, sentence and reply lengths, how often you use contractions, hedges, questions, exclamation marks and emoji, the phrases that recur in your writing, and your language mix. These are numbers and short strings, not text you wrote.
• ›Rules derived from your edits, stored as sentences such as “never open with ‘Hope this finds you well’”. The draft and the message it was compared against are discarded once the rule is derived.
• ›Exemplar references: the Gmail message id of past emails you sent, plus their length, recipient and inferred purpose. The message body is fetched from Gmail at the moment a draft is written and is not retained afterwards.
• ›Your job list: the names you give the things your business runs on, such as a property or a client, plus any aliases you add.
• ›A filing index: for each message Dray files, its Gmail message and thread identifier and which job it belongs to. Identifiers, not content. The filing itself lives in your own Gmail as a label, so it survives whether or not Dray does.
• ›Filing signals: the email addresses of people who correspond about a job, and words from your own subject lines that name it. This is what lets the next message file itself.
• ›Ledger entries you ask Dray to record: an amount, a currency, a counterparty, a date and a one-line description, plus a pointer back to the message it came from. These are read out of an invoice or a quote at your request. The document itself is never copied.
• ›An audit record of every change Dray makes to your mailbox, which you can read at any time.
• ›Your account, plan and billing status.

Message bodies, subject lines, attachment files and calendar data are never written to storage that Saranyx controls. They pass through memory to answer a request and are gone when the request ends. When you ask for the full record of a job, the attachments are listed and fetched from Gmail at that moment; no copy is kept.

The exceptions are deliberate and are listed above: a figure and a one-line description you asked Dray to put in a ledger, and the job names and addresses that make filing work. If that is more than you want stored, do not use the ledger, and delete your account below at any time.

Normally the assistant you are already talking to writes the draft, and Dray only measures it. Nothing extra leaves Google in that case.

If you explicitly ask Dray to rewrite a draft in your voice, it sends the relevant thread, two or three of your own past emails and your style card to Anthropic’s Claude API, which returns the rewrite. Anthropic processes this on Saranyx’s behalf under its commercial terms and does not use it to train models. This is off unless you ask for it.

The host application you use Dray from, normally Claude, also sees the tool results Dray returns. That is inherent to how the Model Context Protocol works.

Dray’s use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements. Specifically, Dray does not transfer Google user data to others except as necessary to provide or improve the product, to comply with applicable law, or as part of a merger or acquisition. Dray does not use Google user data for advertising, and does not allow humans to read it except with your explicit consent, for security purposes, to comply with applicable law, or where the data has been aggregated and anonymised.

Google user data is never used to train any machine learning model, ours or anyone else’s.

This site uses Vercel Web Analytics, which is cookieless. It records aggregate page views and referrers. It sets no cookie, assigns no identifier that follows you between visits, and is never joined to your account, your mailboxes or anything Dray knows about you. The account portal and the connector itself carry no analytics at all.

• ›Cloudflare, which runs the connector and stores encrypted tokens and the audit log.
• ›Vercel, which runs this site and the account portal.
• ›Supabase, which stores accounts, the job list and filing index, style cards and derived rules.
• ›Anthropic, which processes drafting requests as described above.
• ›Stripe, which handles payment. Card details never reach Saranyx.

You can disconnect a mailbox at any time from the Dray portal, or revoke Dray’s access entirely at myaccount.google.com/permissions. Disconnecting deletes the stored refresh token immediately.

Deleting your account is a button in the portal, not a request you have to wait on. It revokes Dray’s access to every connected Google account with Google directly, then deletes every row Dray holds about you: mailboxes, jobs, the filing index, ledger entries, style cards, rules, exemplar references and the audit log. It happens while you wait and it cannot be undone.

Your mail is untouched. The Dray/ labels stay in Gmail, where you can remove them yourself if you want to. If anything is ever unclear, write to privacy@heydray.com.

Saranyx LLC. Questions about this policy go to privacy@heydray.com.