Privacy Policy

1. Introduction

The Proxies AI Pty Ltd (ACN 696 312 473) ("The Proxies," "we," "our," or "us") operates the Proxies platform and related services (collectively, the "Service"). This Privacy Policy explains how we collect, use, disclose, and safeguard your information when you use our Service.

2. Information We Collect

2.1 Information You Provide

• Account registration information (email address, name)
• Authorisation for services you choose to connect, such as Google (see Section 10)
• Profile information you choose to provide
• Communications you send to us

2.2 Automatically Collected Information

• Usage data (features accessed, professions activated, API calls made)
• Device information (browser type, operating system)
• IP address and approximate location
• Cookies and similar tracking technologies

2.3 MCP Protocol Data

When you connect your AI client to Proxies via the Model Context Protocol (MCP), we process the requests your client sends to our tools, and the responses we return, to provide the Service. Conversations in your own AI client stay with that client. Section 3.3 sets out what we do store.

4. How We Use Your Information

• To provide, maintain, and improve the Service
• To authenticate your identity and manage your account
• To process your profession activations and API requests
• To communicate with you about the Service
• To detect, prevent, and address technical issues and security threats
• To comply with legal obligations
• To develop new features and services

5. Information Sharing and Disclosure

We may share your information in the following circumstances:

• Service Providers: Third-party vendors who assist in operating our Service (hosting, analytics, authentication)
• WhatsApp: Meta processes messages and account information to provide connected WhatsApp communications, as described in Section 11.2.
• Legal Requirements: When required by law, court order, or governmental authority
• Safety: To protect the rights, property, or safety of Proxies, our users, or others
• Business Transfers: In connection with a merger, acquisition, or sale of assets
• With Your Consent: When you explicitly authorize us to share information

We do not sell your personal information to third parties.

Support access: Support access is one way our staff can help in your account. You turn it on in your dashboard and choose the named person, which parts of your account they can use, and how long it lasts (up to 30 days). Everything they do is recorded and shown to you in your dashboard, and you can switch it off at any time; it also ends on its own on the date shown. If you allow it, this can include reading, organising and drafting your Microsoft 365 email and calendar, and sending email. Separate OneNote options allow reading pages, creating and editing them, or deleting selected pages and paragraphs. Our team is instructed to make OneNote changes and deletions only on your specific request. It gives our staff no tools to read your Gmail, Google Calendar or Google Drive, and we withhold records in your account that came from Google or may contain Google data (see section 10.4). You can also remove our access to Microsoft at myapps.microsoft.com (work or school accounts) or account.live.com/consent/Manage (personal accounts).

6. Data Security

We protect your information with reasonable security measures. Connections to our website and API are encrypted (HTTPS), and the access keys for Google and the other services you connect under Connections are encrypted at rest. Social provider credentials are stored separately in our publishing service; this statement does not describe encryption of those credentials. Section 10.3 sets out what we store from Google. No method of transmission over the Internet or electronic storage is 100% secure.

7. Data Retention

We retain your information for as long as your account is active or as needed to provide the Service. We may retain certain information as required by law or for legitimate business purposes.

You may request deletion of your account and associated data by contacting us. Some information may be retained in backups for a limited period.

8. Your Rights

Depending on your jurisdiction, you may have the right to:

• Access the personal information we hold about you
• Request correction of inaccurate information
• Request deletion of your information
• Object to or restrict processing of your information
• Request portability of your information
• Withdraw consent where processing is based on consent

To exercise these rights, please contact us at the email address below.

9. Cookies and Tracking

We keep you signed in using your browser's local storage, not cookies, and we store a few preferences and details that keep pages working there too. Our public marketing pages use Google Analytics, which sets cookies to measure how those pages are used. It does not record the pages of your account, such as your dashboard. When it records a page, we remove the extra details that can follow the page address, keeping only the campaign tags that tell us which advertisement or email brought you to the site. You can block or delete those cookies in your browser settings; the Service still works without them.

10. Google User Data

This section explains how The Proxies accesses, uses, stores and shares data from your Google account when you connect it, whether you connect from your own dashboard or as a member of a business that uses The Proxies. Where anything in this section differs from the rest of this policy, this section is the one that applies to your Google data.

10.1 What we access and why

You choose which Google services to connect: Gmail, Google Calendar, Google Sheets, Google Analytics and Search Console. We ask Google only for the permissions of the services you choose, plus your Google account's email address and a confirmation of which account you signed in with, so we know which account is connected, and you can add another service later from your dashboard without disconnecting. When you add an ordinary service to an account, or reconnect it with the Reconnect button your dashboard shows when a connection needs it, we also ask Google again for the permissions that connection already has, so they stay on; an older permission we no longer use, such as Gmail's send-only permission, is not asked for again. Google shows you each permission on its consent screen before you allow it. The optional meeting approval below may ask only for meeting permissions, or ask again for existing permissions, including older ones. We use each one only for the features listed next to it.

• Read Gmail (gmail.readonly): search, list and read your messages and attachments when you or your AI assistant ask; analyse your own sent messages for writing guidance only with your separate workspace setup permission, as described below; read your Gmail signature so drafts carry it; where you connect Gmail as a member of a business, read up to 8 of your most recent sent emails to find the signature you actually use, with its images (such as a logo), and sign the reply drafts we put in your Gmail Drafts with it; show message counts and the messages that need your attention in Email Review; watch your inbox for messages that match automations you set up; and, where your organisation turns them on, check a mailbox connected as a project source and turn incoming emails into proposed tasks; and, if you turn on Email under Live answers for your team on your Proxy-of-me, search your Gmail for messages that match a question from a teammate you have listed (see 10.2).
• Create Gmail drafts and send Gmail (gmail.compose): save, edit and discard drafts in your Gmail, so you can review them before anything is sent, and send a draft once you approve it in Email Review; and send an email when you or your AI assistant ask, and emails sent by automations you have set up.
• Tidy Gmail (optional) (gmail.modify): requested only if you turn on inbox tidy in Email Review, to archive, move to trash or mark as read the messages you select. It is not part of the first connection. If inbox tidy is on, we ask for it again when you add a Google service to that account or reconnect it with the Reconnect button on your dashboard, so it stays on.
• See, create, change, and delete events on Google calendars you own (calendar.events.owned): create and update events on your main calendar: events you or your AI assistant ask for, and meetings confirmed through scheduling pages you share, including moving one that is rescheduled. Where project meeting sync is enabled in your workspace, also match relevant meetings to projects you can update and share their title, time and joining link with internal project members.
• View events on all your calendars (calendar.events.readonly): show your events when you or your AI assistant ask, and see when you are busy, on your own calendars and calendars shared with you, so we can use your availability to suggest meeting times on scheduling pages you share and prepare Proxy-of-me replies about your availability. Proxy-of-me uses busy times by default; if we enable event titles for your Proxy-of-me at your request, it also uses them. Where project meeting sync is enabled in your workspace, also match relevant meetings to projects you can update and share their title, time and joining link with internal project members.
• See the list of Google calendars you’re subscribed to (calendar.calendarlist.readonly): see which calendars you have, so we know which ones to read events from.
• Google Sheets and picked meeting-note Docs (drive.file): read and update only the Google Sheets you pick in Google’s file picker or that we create for you: the spreadsheets you point your AI assistant at, tabs you import into a workbook, an optional one-way copy of ranges you select, a workbook you export to a new Google Sheet, and, where your organisation uses our lead finder, a list of leads from a sheet you choose so a person can review them before any are added to another service you connect, such as Apollo. Where project meeting features are enabled, also read a meeting-note Doc you pick in Google’s file picker and save available notes to a lead or project you can update. Connections made before we moved to this permission may still hold Google’s permission to see, edit, create and delete all your Google Sheets spreadsheets, which we use only for the same features. We no longer ask for it: when such a connection is updated, Google is asked for the narrower permission instead, and each spreadsheet it used has to be picked once.
• Google Analytics (analytics.readonly): read reports for the properties you can access. Read-only.
• Search Console (webmasters.readonly): read search performance and site data for the properties you can access. Read-only.
• Your Google account's email address (userinfo.email): requested with every connection, whichever services you choose, so we know which Google account is connected. We keep it with the connection and show it on your dashboard.
• Confirm which Google account you signed in with (openid): requested with every connection, for the same reason.

Optional project meeting notes. Where this capability is enabled for your workspace, you can pick a meeting-note Doc in Google's file picker and save available notes to a lead or project you can update. For a Doc attached to a Calendar event, connect Google Calendar and Google Sheets first. This uses the same file permission as Google Sheets and does not require the separate meeting approval.

Adding Google Meet conference access requires a separate Google approval on your existing account. It is not included in an ordinary Google connection. The meeting approval may ask only for meeting permissions, or ask again for existing permissions, including older ones. It adds these read-only permissions:

• Read Google Meet conference records (meetings.space.readonly): find completed meetings and their generated transcript or recording references for meetings linked to a project.
• Read files created or edited by Google Meet (drive.meet.readonly): read accessible meeting transcripts and download accessible recordings for transcription and saving to the matching project. This permission does not give us general access to every file in your Drive.
• See all your Google Docs documents (documents.readonly): read shared meeting notes attached to a meeting you link or that matches your internal project or lead. Google grants read access to all Docs you can access; Proxies restricts this feature to those meeting-attached notes and does not scan your other documents.

Optional email writing guidance. In workspace setup, you can separately authorise analysis of your own sent Gmail messages. You choose the connected account, date range and maximum number of messages, up to 500. This uses Read Gmail and adds no Google permission. Saving your choice does not start analysis: you start it in your connected AI assistant (for example Claude).

10.2 How we use it

• We use your Google data only to provide the features above, when you or your AI assistant ask, when an automation you set up runs, or when someone you have allowed asks your Proxy-of-me a question.
• Your AI assistant. When your assistant (for example Claude) calls one of our tools, our servers fetch the data from Google and return it to that assistant so it can answer you. Your assistant's provider then processes it under its own terms. For Claude, that is Anthropic. Anthropic says it does not use raw content from connectors such as ours to train its models, though content copied into your conversation can be included. If you use Claude on a personal plan (Free, Pro or Max), we suggest you keep Model Improvement turned off in Claude's Privacy Settings and not give a thumbs up or down to chats that quote your Google data. Anthropic says it may use a chat to improve its models when Model Improvement is on or when you rate it, keeps a rated chat for up to 5 years, and may also use chats flagged for safety review to improve how it enforces its usage policy, including training the models it uses for that. If you use another assistant, that provider's own terms decide whether your chats are used for training.
• On Claude's commercial plans and its API, Anthropic says it does not use your chats for training by default, unless you rate them or choose to share them.
• Email writing guidance. Your connected AI assistant (for example Claude) receives eligible own-sent samples and prepares private summaries and a proposed writing-style profile. You review the exact proposal on our website. Publishing requires a separate approval that acknowledges sharing with your workspace. Published guidance can be used by helpers, setup sample drafts, ordinary Tasks drafts and revisions. This saves instructions for writing; it does not train a model. Your assistant's provider terms and settings still apply to the samples and conversation.
• Tasks routines. Where your organisation uses Tasks, a routine on your own Claude account can work on tasks for you. Our servers send it the task and excerpts of matching emails from your Gmail, and, where we set it up for your organisation, the text of an incoming email so it can propose tasks. That routine is your AI assistant, so what this section says about your assistant applies to it.
• Features that run on our servers. Automations, agents and meeting scheduling can run without your assistant. The ones that give data from your Google account to an AI model today are isolated Tasks processing, meeting task suggestions, suggested answers to task questions and memory checks, where enabled; meeting scheduling (your busy times) and your Proxy-of-me replies (your busy times by default, plus event titles if we enable them for your Proxy-of-me at your request, and matching messages from your Gmail if you turn on live answers for your team, described below). Event descriptions and locations are not included in this calendar context. Where project meeting transcription is enabled, accessible recordings also go to a speech-to-text model on servers we operate ourselves. Apart from that recording transcription, when a feature that runs on our servers needs an AI model to work with data we retrieved from your Google account, that data is processed only through Anthropic's API, and never by any other AI provider, except for the search vectors and the image and video requests described below. Anthropic says it does not use data sent through its commercial API to train its models unless the customer chooses to share it. We do not ask Anthropic to keep this processing in one country, so Anthropic may run it in any of the regions it uses (see Section 13). Where isolated Tasks processing is enabled, our servers also send task context and relevant Gmail content to Anthropic’s API to prepare drafts for your review. Email-to-task analysis requires your separate approval for this server-side processing. Where meeting task suggestions are enabled and you ask for suggested tasks from the notes or transcript of a meeting linked to a project or lead, our servers send that text to Anthropic’s API, and nothing becomes a task until you confirm it. Where suggested answers to task questions are enabled, our servers send a task’s title, any notes typed on the task in our app, and parts of its routine work report, which can come from your email, to Anthropic’s API, using your organisation’s own Anthropic API key where it has added one. Those parts are the report’s understanding of the task, its summary, findings and open questions, and the reasons a draft is held before sending. Where memory checks are enabled, when your assistant or a routine saves a memory, our servers send it and similar saved memories to Anthropic’s API to decide whether it replaces an older one. A memory can quote your Google data. Companion scans the inbox of each mailbox it is turned on for (all connected mailboxes unless you choose specific ones) and may also search those mailboxes and your tasks for context, including messages outside the initial scan’s folders and dates. If we later move this processing onto AI servers we operate ourselves in Australia, everything in this section will still apply to it.
• Project meetings. Where Projects & CRM and project meeting sync are enabled, we use your connected calendar to match relevant meetings to projects you can access and update. We skip events marked private or confidential and meetings with an ambiguous project match. Your whole calendar is not shared with the team. A matched project can show the meeting title, time and joining link. After a meeting, if you have approved Google Meet notes access, we automatically save accessible generated transcripts to the matching project. If Google has no generated transcript, an accessible recording can be transcribed on servers we operate ourselves and the transcript saved there. You can also link a meeting and save available notes manually. People with internal access to the project can read saved notes, and your helper can retrieve them to answer questions or prepare follow-up work. Google must have produced and made the source available to your account; connecting Proxies does not start recording or transcription in Google Meet.
• Live answers for your team. In a business workspace on The Proxies, if you turn on Email under Live answers for your team on your Proxy-of-me, then when a teammate asks it a question, our servers search the Gmail you connected in that workspace for messages that match the words of the question. This happens only for teammates who are signed in to that workspace and whom you have listed one by one under Authorised people with the team tag. We read only each message's sender, date, subject and Gmail's short preview, never the rest of the message. Up to five matches from the last 12 months, with previews cut to 200 characters and links removed, go to the AI model that writes the reply, which is only Anthropic's Claude, as described above. Only the teammate who asked sees the reply, and you can read it under Conversations. Messages whose sender, subject or preview names a topic you have marked Hidden in your Notes are left out. We do not keep the messages themselves (see 10.3), and, as with all your Google data, we never use them for advertising or to train AI models. Email is off until you turn it on, and you can turn it off at any time on your Proxy-of-me's About you page.
• Workspace onboarding. Our guided workspace setup (Bodhi) uses Anthropic’s Managed Agents API to process the public website pages, instructions and uploaded source text supplied for that workspace. Bodhi does not connect to your Google account; if you supply Google-derived content yourself, it is processed in the same way. We retain the onboarding conversation and source files so you can continue setup and correct information. Readable source copies are held in Anthropic’s file storage and in the onboarding conversation until we delete them. Skipping a file does not erase these copies. Email privacy@the-proxies.ai to request deletion.
• Our search and knowledge tools. OpenAI's API turns text into search vectors so we can find matching documents, or add text to a knowledge base when you upload a document or your assistant asks us to. That text is what your assistant sends to our search and knowledge tools or to a helper that searches a knowledge base, and the title of a task, including one created from your email, when we search a knowledge base to prepare that task for your assistant. Where isolated Tasks processing is enabled, it also includes the searches and helper questions that processing writes while it prepares a draft, which can draw on your Gmail. It also includes the text of documents you upload to Knowledge, including when we rebuild the index from retained documents after a source is removed. Bodhi onboarding also sends the workspace knowledge it prepares and the questions it uses to check retrieval to OpenAI for search vectors. If that text includes content from your Google account, OpenAI uses it to make search vectors, and may keep it for up to 30 days to monitor for abuse, or longer where the law requires it or to prevent harm. OpenAI says it does not use data sent to its API to train its models unless the customer chooses to share it.
• Images and videos. Where image or video creation is enabled, your assistant can ask our tools to create an image or a video. Our servers then send Replicate's API the description your assistant writes and, for a video, any starting image it gives. That description can come from Google data your assistant has read. Replicate says it removes the inputs, outputs and logs of requests made through its API after an hour by default. Its terms let it use them only as far as needed to provide the result and its services, to train models that a customer fine-tunes through its service, which we do not do, and to compile aggregated, anonymised data. Its privacy policy says it may keep personal information longer where the law requires it or for its legitimate interests, such as preventing abuse, and in backups for a limited period.
• We do not use your Google data to train, create or improve any artificial intelligence or machine learning model.
• We do not sell your Google data, use it for advertising, or use it to decide creditworthiness or lending.

10.3 What we store and for how long

• Email writing guidance. We keep your saved analysis permission, progress, message references and fingerprints, private summaries and proposal versions. This learning record stores references rather than raw message bodies, but summaries supplied by your assistant can still include personal information or quotations. Published guidance and its approval and recovery history are retained separately. These records have no automatic expiry. Withdrawing analysis permission removes the private learning journal and source evidence and rejects unpublished proposals; it does not erase proposal history, published guidance or copies held in your assistant conversation. Eligible withdrawn private history can be deleted in workspace setup. History supporting published knowledge is protected there; request its deletion separately using 10.6.
• Access keys. The tokens Google issues when you connect, encrypted at rest (AES-256-GCM). Deleted when you disconnect. We also keep the email address of the Google account you connected (and its name and picture, where Google provides them), so you can tell your accounts apart. We also keep the ID Google gives that account, so we can tell when two connections are the same Google account. It is erased when you disconnect.
• Calendar events. If you connect your calendar, we keep a synced copy of your events (times, titles, descriptions and locations) so we can check your availability. For a calendar connected through Connections, the copy is deleted when you disconnect. For a calendar you shared with us instead, removing it from your calendar page in The Proxies deletes the copy.
• Saved project meetings. Meeting cards and saved notes are stored with the project. Removing a meeting link stops its automatic recreation but does not delete saved notes. Disconnecting Google stops new reads; it does not erase records already shared with a project. Saved notes remain until deleted from the project. Email privacy@the-proxies.ai to request erasure of saved project notes, their history messages and meeting cards. Temporary recording copies use private storage and are deleted after processing or expiry; private transcription results expire after seven days. Job records, including the recording file name, remain for up to 30 days. A transcript saved to the project remains as a project record.
• Inbox automation alerts. If you set up an automation that watches your inbox, we keep the subject, sender and a short preview of matching messages for up to 30 days, then erase them, keeping only each message's identifier so we do not alert you twice.
• Automation send records. When an automation sends an email through your Gmail, we keep the recipient and subject as a record of what was sent on your behalf. On accounts where we have turned on send logging for quality checks, and for emails addressed to our own team, we also keep the email body. We delete the recipient and subject record after 12 months, and any stored email body after 30 days.
• Email Review approvals. When you approve a draft for sending, we keep a record of the approval with message identifiers and fingerprints (hashes) of the content. We do not keep the email itself or its recipients. The record stays after sending so the same draft cannot be sent twice. When you remove the account or ask us to delete your data, we erase the separate fingerprints of the subject, recipients and body. Each record keeps the draft's identifier and one fingerprint of the whole draft, which is what tells us a draft was not altered before it was sent, and the message identifier of a draft that was sent or may have been sent.
• Spreadsheet data you import. Tabs you import from Google Sheets into a workbook, and ranges you choose to keep in sync, are stored in that workbook. Deleting a workbook only hides it: its data and history stay in our database until your organisation's data is destroyed. Email privacy@the-proxies.ai if you want a deleted workbook erased.
• Tasks proposed from email. Where your organisation turns this on, a task proposed from an incoming email keeps a title, a short description and an excerpt of the email of up to 1,200 characters. It is kept with the task. During isolated processing, we also store working context, tool results and model responses so interrupted work can resume. These working copies can contain more email text than the task excerpt. Prepared drafts and analysis results may also contain email-derived content.
• Contacts added from email. Where we turn this on for your organisation, a routine that reads your email can propose adding a person introduced in a message to your organisation's Projects & CRM. Our servers read that message again in your Gmail to check the proposed name and email address, and we keep each proposal, the result of that check and what happened next with that routine's saved notes. Where we turn on adding contacts, our servers then add the contact, a lead where the routine asks for one, and, where the routine names who introduced them, a note on that person's record. In a trial mode, they only check your CRM and record what they would add, and change nothing in it. Other proposals wait for a person to review them. A contact and lead that are added stay in your CRM, visible to people who can see it.
• Lead lists. If your organisation uses our lead finder, the lead details read from the sheet you choose (names, organisations, email addresses, phone numbers and websites) are stored so a person can review them.
• Email signature. Where you connect Gmail as a member of a business, we keep a copy of the signature we found in your recent sent emails, with its images, on that connection, so we do not read your sent emails for every draft. We refresh the copy from your sent emails when it is more than a week old. If you edit the signature and save your own version, we keep that version instead. Choosing Turn off next to the account under Connections stops us adding it to your drafts. The copy is deleted when you disconnect the account or ask us to delete your data.
• The record of the connection. Where you connect as a member of a business, removing the account or asking us to delete your data switches the connection off and erases its tokens, and we keep an inactive record of the connection holding your Google account's email address, and its name and picture where Google provides them. Email privacy@the-proxies.ai if you want that record erased as well. If the business closes its account with us, that record is deleted too. A Google account still connected there at that point is disconnected first, and we ask Google to remove our access to it only when we can confirm that no other connection in The Proxies uses that account.
• Live answers for your team. We do not keep the Gmail messages your Proxy-of-me reads to answer a teammate. Its reply, which may mention them, is kept with that conversation like every Proxy-of-me reply, together with a count of how many messages it used. Email privacy@the-proxies.ai if you want a conversation deleted.
• Mailbox checks. If your organisation asks us to connect a mailbox as a project source, we check the connection by recording only a count of recent messages and the dates of the oldest and newest. These records are kept for 90 days.
• Routine work reports. Where your organisation turns this on, a routine that researches a task for you stores what it found, references to the messages it used and any draft emails it prepared. The report is kept with the task.
• Workspace onboarding. We keep original onboarding uploads in Cloudflare R2 or our database, along with extracted text and the conversation, so setup can continue. These records have no automatic expiry: they remain until we complete a deletion request or delete the organisation’s workspace. The database copies are included in the encrypted backups described below. Knowledge prepared from these sources is also stored in our database and search index. Deleting an onboarding conversation does not remove knowledge already published to your organisation’s workspace; request deletion of that knowledge separately.
• What you ask your assistant to save. If you ask your assistant to remember something, create a task or save a file, what it saves may include information from your Google account. We treat it as your own content, and you can delete it. When you upload a document to Knowledge, we also keep the original file in our document storage (Cloudflare R2) and its extracted text to support retrieval and recovery. Removing a document from Knowledge removes it from helper search, but does not remove the retained original file or its extracted text. Follow Section 10.6 to request deletion of both.

Apart from the items above, we do not keep copies of your email messages, attachments, calendar or files. The Google data we store in our database is also kept in our encrypted backups of that database. We delete each of those backups once it is more than 30 days old, so data you delete can stay in them for about 30 days. Text from documents you upload to Knowledge or ask your assistant to add to a knowledge base, including the text we keep to rebuild the index, is stored in our search index and our database, and can stay in that index's backups for longer.

Our own team's mailboxes. We also use this Google connection on our own staff Gmail accounts. Every hour, our servers search them for meeting-notes emails from Google Meet (Gemini) and from Plaud, and save each new note to our internal team records, kept in our private GitHub repository: its title, date and text, any link to the full notes and, for Google Meet notes, the email's recipients. This reads only those staff accounts, never yours. Our founders and administrators can also use their own Gmail and Calendar in an internal console, where Anthropic's API works with that data and can ask Replicate for an image. A routine also reads a founder's own Gmail to keep our own customer records up to date, using Anthropic's API, and saves short excerpts in those records.

10.4 Who can see it

Our staff do not read your Google data, except:

• where you have given explicit consent for us to see specific data, for example when you ask us to help with a problem and agree to share it;
• where it is necessary for security, such as investigating a bug or abuse;
• where it is necessary to comply with the law; or
• where it has been aggregated and anonymised for our internal operations.

For example, our staff use our support tools to open a workbook holding data imported from your Google Sheets only when you ask us for help with it, and those tools keep a record in your organisation's activity log when they do. If you choose to send us a conversation as feedback and ask for it to be emailed, it reaches us through our email provider, Resend.

Support access, which you can turn on in your dashboard, gives our staff no tools to read your Google data and withholds records that came from it; it can only create Gmail drafts or send new emails that you allowed.

10.5 Sharing

If you separately approve publication of email writing guidance, it becomes shared workspace knowledge. Personal writing-style selection by a helper does not make that shared record private. Workspace members, their AI assistants and drafting routines can receive applicable published guidance.

We share Google data only:

• with the AI assistant you connect, when it calls our tools or when we send the data described in 10.2 to a Tasks routine on your own Claude account;
• with members of your workspace and their AI assistants and drafting routines, for email writing guidance you publish;
• with people who open a scheduling page or Proxy-of-me page you share, who see times you are free, worked out from your calendar, and replies about your availability drawn from it;
• with teammates you list for live answers on your Proxy-of-me, who see replies that can draw on matching messages from your Gmail (see 10.2);
• with people who have internal access to a project, who can see its linked meetings and saved notes as described above;
• with the service providers that host or support the features above (Fly.io, Cloudflare, GitHub, Upstash and Sentry, all listed in Section 11; GitHub runs our encrypted backups, and Anthropic, OpenAI and Replicate are described in 10.2);
• with our email provider, Resend, for emails our automations send, scheduling emails such as meeting invitations and confirmations, task notifications to members of your workspace, which can include the title of a task proposed from your email, and feedback you ask us to email;
• with our search-data providers DataForSEO and Serper, when your assistant sends them search terms, which can come from any Google data it has read through our tools or that we sent it, such as your Search Console data or your email;
• with another service you connect, such as Apollo, when a person approves a lead for it;
• with other services you connect yourself, such as Hunter, Mailchimp or Notion, when your assistant sends them searches or other text through our tools;
• with public search and reference services, such as GDELT news search, MET Norway weather forecasts, the image libraries Unsplash, Openverse, Wikimedia Commons and Trove, and address and city map services, when your assistant sends them search terms, places or addresses through our tools or cards;
• with a currency reference rate service, which receives only a currency code and a date;
• with Replicate, where image or video creation is enabled, when your assistant asks our tools to create an image or video from a description it writes (see 10.2);
• where your organisation turns on the task mirror, with a private GitHub repository it chooses, which receives task titles, descriptions and history, including tasks proposed from your email, and where copies cannot be fully erased;
• to comply with the law;
• for security; or
• as part of a merger, acquisition or sale of assets after asking for your explicit consent.

What your assistant sends to other services you connect, to public search and reference services and to Replicate can come from Google data it has read. By default, people who open a scheduling page or Proxy-of-me page you share do not see the titles, descriptions or locations of your events. If we enable event titles for your Proxy-of-me at your request, its replies to them may also include event titles.

10.6 Disconnecting and deleting your data

To stop email-style analysis, choose Withdraw permission under Learn your email style in workspace setup. Delete withdrawn history removes eligible private history, not published workspace knowledge. Disconnecting Google or withdrawing analysis permission does not stop helpers and drafting routines from using guidance already published. Manage that guidance separately, or email privacy@the-proxies.ai to request deletion of the guidance and its retained history. Copies in your AI assistant account follow that provider's deletion controls.

• Disconnect. In your dashboard, find the Google account under Connections and choose Remove. We delete the access keys we hold, the calendar events we copied from that account and the copy of your email signature. We ask Google to remove our access to that account only when we can confirm that no other connection in The Proxies uses it. Removing the connection does not delete your saved email-analysis permission, message references and fingerprints, progress, private summaries or proposals. Use Withdraw permission and Delete withdrawn history as described above; published guidance and its protected history need a separate deletion request.
• Remove our access at Google. When you remove a Google account, we ask Google to remove our access only when we can confirm that no other connection in The Proxies uses that Google account, and Google can take a while to apply it. If we cannot confirm it, we do not ask, and you can remove our access at Google yourself. To check, or to remove it yourself, go to your Google Account's third-party connections page, select The Proxies and remove access. Removing it there stops every connection to that account in The Proxies. If you shared a calendar with us, also remove that share in your Google Calendar settings.
• Ask us to delete your data. Email privacy@the-proxies.ai. A deletion request removes the access keys for your connected accounts, the calendar events we copied, the copies of your email signature, your inbox automation alerts, the memories and files you saved, and onboarding conversations and source copies (including Anthropic copies). You can also request erasure of saved project meeting notes, their history messages and meeting cards. For each of those Google accounts, it asks Google to remove our access only when we can confirm that no other connection in The Proxies uses that account.

10.7 Limited Use

The Proxies' use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. The use of information received from Google Workspace scopes will adhere to the Google User Data Policy, including the Limited Use requirements. The use of raw or derived user data received from Workspace APIs will adhere to the Google User Data Policy, including the Limited Use requirements. See the Google API Services User Data Policy and the Google Workspace API User Data and Developer Policy.

11. Third-Party Services

We name the third parties that handle data on our behalf rather than describing them in categories. Our current subprocessors are:

• AI providers. Anthropic (the primary model behind the helpers), OpenAI (used only to turn documents and search requests into search vectors for our search index, never to generate content; Section 10.2 explains when that text can include content from your Google account), and Groq, which serves one server-side feature, turning a driver's voice note into a list of the items they collected, and never receives data from your Google account. Replicate generates images and video where image or video creation is enabled and a task requires it, and reads the text of scanned documents and photos you upload (optical character recognition).
• Infrastructure. Fly.io (compute, database), Cloudflare R2 (document storage and backups), Tigris (encrypted Social database backups through Fly.io object storage), Upstash (caching), GitHub (source control, deployment and running our encrypted backups), and Sentry (error reports from our servers). Our main API server removes request details, user information and free-text content from these reports, keeping diagnostic categories, counts, code locations, routine and routine-run references, and a hashed workspace identifier for grouping. These identifiers are not anonymous. Reports from other services may still include details of the request that failed.
• Billing and email. Stripe for payments, Resend for transactional email, and MailerLite for marketing email you have opted into.
• Other services. Serper and DataForSEO, used by your assistant, specific helpers and agents to retrieve public web and search data. The search terms your assistant sends them can come from any Google data it has read through our tools or that we sent it, such as your Search Console data or your email (see Section 10.5). We also use Google Analytics, which measures how our public marketing pages are used (see Section 9).
• Services you connect yourself. Where you authorise a connection to Google, Microsoft 365, WhatsApp (Meta) or a similar provider, we retrieve data from that service on your instruction. Section 10 sets out what we store from Google and for how long; Section 11.1 covers Social and Section 11.2 covers WhatsApp.

These third parties have their own privacy policies, and we encourage you to review them. We maintain a fuller subprocessor register and provide it on request. If we add or change a provider that handles your data, we will tell you.

11.1 Social accounts and publishing

Where Social connections are available, you can authorise Facebook Pages, Instagram professional accounts and LinkedIn profiles or Pages. The provider shows the permissions before you agree. We use account and Page identifiers, names, profile images and access credentials to identify the accounts you choose and carry out publishing you approve. Our publishing service stores the provider access credentials. Provider approval and your account permissions determine which connections are available.

We store your Social drafts, uploaded media, account connections, approvals, schedules and publication results to provide Social in your workspace. Our hosted publishing service sends approved content to the account you select. Connecting an account does not approve a post: each post needs approval before it is scheduled. If you ask your AI assistant to work with Social content, the information returned to that assistant is processed by its provider under its own terms.

Media submitted for approved publishing is made available at an unguessable public web address so the social network can fetch it. Anyone with that address can retrieve it; it is not a private file-sharing link. Connected accounts' public profile images may also be stored there. Primary Social publishing records are hosted in Sydney, Australia, and this media is stored with Cloudflare R2 using its Oceania location setting. Encrypted Social database backups are stored with Tigris; we do not guarantee their storage in Sydney. Fly.io also stores encrypted volume snapshots in the United States. Network-edge processing and support access may occur outside Australia, as described in Section 13.

You can revoke access in the social provider's account settings. Revoking access does not remove content already published on that network or erase records we retain. To request deletion of your Social connection and associated data, follow our data deletion instructions. We verify the request and its scope before removing data. Some records may be retained for legal obligations, and backups expire under their retention schedule. Copies already held by a social network remain subject to that network's policies.

11.2 WhatsApp business messages

For accounts connected to Connect through Meta’s official WhatsApp Business Platform, we process account and telephone identifiers, participant telephone numbers, message content, timestamps and delivery status to provide its authorised Connect inbox. Where enabled, messages may include attachments such as photos, documents and voice messages. Meta processes WhatsApp communications under its own terms and privacy policies. This section describes the official Connect integration; it does not describe legacy linked-device connections.

Credentials for these official Connect accounts are encrypted at rest. Access to saved conversations and attachments is restricted to authorised workspace members. If you ask an AI assistant to process message content, the information supplied to that assistant is also processed by its provider under its own terms.

Disconnecting WhatsApp stops connected access but does not erase messages, attachments or other records already saved in your workspace. Deleting operational delivery records does not delete these saved copies. To request erasure, follow our data deletion instructions and identify the workspace and connected WhatsApp number. We verify your authority and the scope of the request, including any saved messages and attachments. We will explain applicable legal and backup retention when responding. Copies held by Meta or other message participants remain subject to their own controls and policies.

12. Children's Privacy

The Service is not intended for users under 18 years of age. We do not knowingly collect information from children. If you believe we have collected information from a child, please contact us immediately.

13. International Data Transfers

We are an Australian company and we tell you where your data actually sits rather than leaving it implied.

• Our primary account database, search index, knowledge store and Social publishing records are hosted in Sydney, Australia. This does not mean that every backup, network request or support operation stays in Australia.
• Uploaded documents and our Cloudflare offsite backup copies use Cloudflare R2's Oceania location setting. This setting is not a guarantee that all processing or access is confined to Australia. Encrypted Social database backups are stored with Tigris without a guaranteed Sydney storage location.
• Fly.io stores encrypted volume snapshots and control-plane data, including service configuration and secrets, in the United States. Its network-edge processing and authorised staff support access can occur in other countries, regardless of our Sydney hosting region.
• External services, including Meta for WhatsApp and AI providers listed in Section 11, may process information outside Australia. Their processing locations and access arrangements depend on the provider and service. If we run AI models on our own servers, they will be in Australia.

Backups, infrastructure operations and external processing may transfer information you submit to a country other than Australia for storage or processing, where different data protection laws may apply. By using the Service you consent to that transfer. If you have a residency requirement, for example that your data must remain in a specific country, contact us and we will scope the arrangement with you.

14. Changes to This Policy

We may update this Privacy Policy from time to time. We will notify you of material changes by posting the new policy on this page and updating the "Last updated" date. Your continued use of the Service after changes constitutes acceptance of the updated policy.

15. Contact Us

If you have questions about this Privacy Policy or our data practices, please contact us at:

Email: privacy@the-proxies.ai