Privacy Policy

Collectly AI is an accounts-receivable assistant. It reads a business’s own invoices and the replies its customers send to Collectly AI’s follow-ups or that the business pastes in, works out what is happening with each unpaid invoice, and drafts follow-up emails for a person to approve and send. A business can also turn on automatic sending, which sends routine low-risk reminders without that approval.

This policy describes what data Collectly AI handles, why, where it goes, how long it stays, and how to remove it. It is written to be checked against the software’s actual behaviour rather than to sound reassuring.

Almost every hard question about a product like this comes from confusing these, so they are separated before anything else.

• You, the business using Collectly AI. You signed up, you chose to connect a mailbox, and this policy is our agreement with you about your own account data.
• Your customers, whose replies Collectly AI reads. They did not sign up for anything and were never shown this page. Their replies are in Collectly AI because they answered a follow-up you sent through it, or because you pasted them in. For that data you are the controller and Collectly AI is your processor: we handle it on your instructions, for the purpose of chasing your invoices, and for nothing else.

If you are one of those customers and you have found this page: the business that emails you holds the record and decides what happens to it, so ask them first. You can also write to us at the address at the bottom, and we will pass it on or act on their instruction.

Connecting a Gmail account asks Google for four permissions. One of them is a Gmail permission, and it is the one that does the work:

• gmail.send — to send a follow-up after you have reviewed and approved it, from your own address, replying into the existing thread where there is one. If an owner or admin of your workspace has turned on automatic sending for low-risk reminders, it is also used to send routine low-risk reminders the same way, without anyone approving each one. That setting is off unless someone turns it on. It is also used to email a team invitation, from the same address, when an owner or admin of your workspace invites someone to join it.

The other three are Google’s standard sign-in scopes: openid, email and profile. They are not Gmail permissions and give no access to mail. They are how the connection knows which Google account just authorised it, so the mailbox is recorded against the right address.

Sending is the one thing Collectly AI adds to your mailbox: each message it sends with gmail.send appears in your Sent folder, as any email you send does. Collectly AI does not label, archive, modify or delete your existing messages, and it does not create or manage Gmail drafts. Follow-up drafts are held inside Collectly AI for your approval, not written into your Gmail drafts, so no broader scope is needed.

Collectly AI has no permission to open or search the mail in your mailbox. What it receives from Google is the access tokens that let it send, the address and name of the account you connected, and, each time it sends, the identifiers Gmail gives the sent message and its conversation, which it uses to send the next follow-up in the same conversation.

If you connected Gmail before 19 October 2026, your Google account may still list a broader permission that Collectly AI asked for then. Collectly AI no longer asks for it or uses it. Disconnecting and connecting again replaces it with the send permission alone.

Replies do not come from Google. Each follow-up Collectly AI sends carries, next to your own address, a reply address on a Collectly AI subdomain that is unique to that message. A customer who replies writes to both: the reply reaches your inbox as usual, and our email provider, Resend, receives it at that address and passes it to Collectly AI.

• Collectly AI only sees replies sent to its follow-ups. Anything a customer sends you another way stays in your inbox unless you paste it into an invoice.
• A reply is read only if it comes from an address on file for that customer and Resend's receiving server confirms the sender with SPF, DKIM or DMARC. Anything else is held back: the invoice shows who it came from and why it was held, and nothing else from it is stored or read.
• A reply that is read is stored and analysed exactly like one you paste in, as described under What we store and Who else sees this data below.

Specifically, Collectly AI does not:

• sell Google user data, ever;
• use Google user data for advertising, profiling, or any purpose beyond providing the receivables features described here;
• allow any person to look at your email or your customers' replies, except where you explicitly ask us to for support, where required by law, or to investigate a security incident;
• use Google user data to train generalised machine-learning models.

Collectly AI does not use Google user data to develop, improve or train AI or machine-learning models of any kind.

Email content is sent to Anthropic’s API to be analysed and to draft a reply, which is the feature you are using Collectly AI for. That transfer is described in full below.

• Email content. The follow-ups Collectly AI sends, and the customer replies that reach it or that you paste in, are stored so Collectly AI can show you the conversation and reason about it. That includes sender, recipients, subject, timestamp and body text, up to 20,000 characters per message.
• Access tokens. Your Gmail refresh token, and any QuickBooks or Xero token, is encrypted with AES-256-GCM before it is written to the database, decrypted only inside the server module that calls the provider, and never sent to a browser.
• Invoices and customers. Whatever you enter, import from CSV, or sync from a connected billing system — including a customer contact name and phone number if you fill them in.
• Account data. Your name, email address, a bcrypt hash of your password, whether and when you proved control of the address, and a profile image URL if you signed in with Google. Signing in with Google stores no Google token: only the name, email address and profile image URL that Google returns are kept.
• Security audit records. Each audited action — a sign-in, a failed sign-in, an invitation, an export, a disconnect — is recorded with the acting account, the time, the IP address the request came from and the browser user-agent string. These records never contain message contents, amounts or tokens.
• Signed-in devices. Each sign-in is recorded as a session: when it started, when it was last used, when it expires, and whether and why it was signed out, together with the IP address it signed in from and the browser user-agent string, cut to 200 characters. This is what the list of signed-in devices in Settings shows, so that you can recognise each device and sign it out.
• Rate-limiting counters. To stop abuse of endpoints that need no account, short-lived counters are kept against an identifier — the account for a signed-in action, the request’s IP address for a public one.
• Onboarding survey answers, if you give them. A new workspace is offered four optional multiple-choice questions — the kind of work you do, what you came to Collectly AI for, roughly how many invoices are unpaid at once, and how you heard about us — plus a short optional free-text box. It is skippable, skipping it costs you nothing, and the answers are used only in aggregate to understand who Collectly AI is for. They never change how your invoices are handled.
• Monthly feedback answers, if you give them. Once a month, in the two days before your subscription’s monthly billing date, an owner or admin of a workspace on a paid plan may be asked how likely they are to recommend Collectly AI, on a scale of 0 to 10, and what would make it more useful. The comment is optional. We store the score, the comment as typed (up to 2,000 characters), who answered, for which workspace and which month, and when. Choosing “Not now” is stored too, with no score or comment, so that the question is not asked again that month. The answers are read by us, on an internal page, to see how Collectly AI is doing and what to change; they never change how your invoices are handled and are never sent to a language model. They have no automatic expiry: an answer is kept until the person who gave it deletes their account or the workspace is deleted, and is deleted then. Your own answers are included in the copy of your data in Settings.
• Support messages. What you write to us through the product, so that we can answer it.

Five parts of this site work without an account. They are listed here because a policy covering only signed-in users is not a complete one.

• The reply decoder. Text pasted into the reply decoder is sent to the language model that answers it and is not stored: not in the database, not in our logs, not in any error report. What is kept is a one-way SHA-256 hash of the text, the category it was classified as, and the time — enough to count how often the tool is used, and not enough to reconstruct a word of it. A test in the repository pushes a marker string through the whole endpoint and fails if it reaches any database column.
• The mailing list. If you ask for a template pack or a teardown from one of the public calculators, we store your email address, which offers you asked for, which page you asked from, the IP the request came from, and the confirmation and unsubscribe tokens. Nothing is ever mailed to that address until you click the confirmation link, and the unsubscribe link in any message keeps working permanently. The list is kept deliberately separate from accounts: asking a calculator for a PDF is not signing up for Collectly AI, and we do not treat it as though it were.
• Applying to sell Collectly AI. There is an unlisted page, sent to people individually rather than published, describing a commission arrangement for selling Collectly AI and carrying a form. If you fill it in we store your name, your email address, your phone number if you chose to give one, and your three short answers — and nothing else. No IP address, no browser, no referrer. It is held in its own table, separate from accounts and separate from the mailing list, is read only by the operator behind a sign-in, is never added to any mailing list, and is never sent to a language model. There is no automatic expiry on it: unlike the periods listed below, nothing deletes an application on a schedule, so it stays until it is deleted by hand. The operator can do that from the page they read applications on, and it takes effect at once — the record itself is destroyed, and what remains is a log entry saying that an application was deleted, by which operator and when, carrying nothing about you. Ask at hello@collectlyai.org and it will be.
• Following a partner’s link. Some people who recommend Collectly AI are paid a commission for it, and each has a share link ending in /r/ and a code. Opening one adds one to a count kept for that partner for that day, and the count is all that is recorded — no IP address, no browser, no referrer, no time finer than the day, nothing that describes you. Link previews that Slack, iMessage, WhatsApp and similar services build when a link is pasted are recognised by the name they send and not counted; that name is read to decide and is not kept. The IP address is used only for the short-lived rate-limiting counter described above. Opening the link also sets the referral cookie described in the cookie policy, and if you then create an account, your workspace is recorded as referred by that partner. The daily counts have no automatic expiry: they are kept for as long as that partner’s record is.
• Being a referral partner. For each partner, Collectly AI holds the name and email address the operator entered, the operator’s own notes, and a private link to that partner’s dashboard — of which only a SHA-256 digest is stored, so the link cannot be read back out of the database, and issuing a new one replaces the old one rather than revealing it. Opening the dashboard records nothing beyond the rate-limiting counter described above. None of this expires automatically, and there is not yet a button that deletes it; ask at hello@collectlyai.org and it will be deleted by hand.

A nightly job removes the following once the stated period has passed. It clears fields; it does not delete invoices, customers, follow-ups, or the record that something happened, because those carry accounting and audit significance. The periods are configurable per deployment, and these are the values in use.

• Email message bodies — 365 days. The short preview line survives, because it is what the timeline shows. Two deliberate exceptions: a body is not removed while its invoice is still open, disputed, or flagged for human review, because the exact words are the evidence in a live argument. The consequence, plainly — an invoice never paid and never written off keeps its correspondence indefinitely, and writing off a bad debt is what releases the clock.
• The model’s raw response — 30 days. A second copy of the same correspondence, kept for debugging and shown on no screen. The assistant’s summary and reasoning survive, because they are the record of why a follow-up was sent.
• IP address and user-agent on audit records — 90 days. The record of the action survives; the fields that describe the person rather than the event do not. That includes an IP or a hashed sign-in address recorded inside the detail of a record, not only the dedicated fields.
• IP address and user-agent on signed-in device records — 90 days. Counted from sign-in. A session lasts at most 30 days, so by then it has long since ended. The rest of the record is deleted outright 90 days after the session expires.
• The IP recorded when someone joins the mailing list — 30 days.
• The free-text onboarding note — 365 days. The multiple-choice answers stay.
• Timeline entry detail — 730 days. The title of each timeline entry survives; the description under it, which can carry a fragment of a customer’s message, does not.
• Spent password-reset requests — 90 days. A reset request records the IP it came from. The link itself is good for an hour; the record of the request is kept for the same window as the other security records and then deleted outright.
• Invitations that can no longer be accepted — 90 days. An invitation holds the address it was sent to, which often belongs to someone who never signed up. Once it has been accepted, revoked or has expired, it is deleted; who is actually in a workspace is recorded separately.
• Rate-limiting counters — deleted as soon as they expire. The counters that slow down password guessing and sign-up abuse are keyed on an IP address or an email address. A counter is of no further use the moment its window closes, so the nightly job deletes it then. The longest window is 24 hours.
• Why a digest email could not be sent — 90 days. When the mail provider refuses your digest email, the reason is kept on your membership, with any email address in it masked, so Settings can show it. The next digest that goes out clears it; otherwise it is removed after 90 days.

Everything else — invoices, customers, follow-ups, the fact that an action happened — is kept while your account exists and goes when it is deleted. Everything named above is reached by the nightly job. Where that was not true this page said so; it is true now.

Collectly AI uses the following processors, and this is the complete list. Each receives only what is described.

• Anthropic — email content, invoice details and the business context you write in settings are sent to Anthropic’s API so the assistant can analyse a situation and draft a reply. This is the core of the product and cannot be switched off while the assistant is in use. Under Anthropic’s commercial terms, API inputs and outputs are not used to train its models. Requests go directly to Anthropic’s own endpoint; the code pins it, so no hosting provider’s AI gateway can sit in between.
• Google — the Gmail API, on the grant you gave it, to send the follow-ups you approve, along with any routine low-risk reminders sent automatically if your workspace has turned that on, and the team invitations described above.
• Neon — managed PostgreSQL, where everything above is stored.
• Netlify — application hosting. Netlify handles every request and keeps its own request and function logs, on its own schedule.
• Resend — the mail provider for messages Collectly AI sends about your account: password resets, address verification, subscription renewal notices, support replies and the digest email (what is waiting for you in a workspace, which names the customers and amounts involved, sent only to members of that workspace and switched off in Settings or from the email itself); and for the mailing list, its confirmation messages and, once you have confirmed, the messages sent to it. Resend receives the recipient address and the message text. Follow-ups to your customers are not sent through Resend — they leave your own mailbox, through Gmail — but your customers’ replies to them reach Collectly AI through Resend: it receives each reply sent to a Collectly AI reply address, with its sender, recipients, subject, headers and text, and passes it to Collectly AI. Resend keeps received mail for 30 days under its own retention policy.
• Stripe — subscription billing. Stripe never receives your email content, and card details never reach our servers.
• Intuit and Xero — only if you connect them, and only to read invoices. Collectly AI never writes to your accounting system.

One more party sees a narrow slice of account data, and is not a processor. If you created your account through a referral partner’s link, that partner can see, on their own dashboard, a short reference standing for your account, the date it was created, the date of its first payment, the plan and billing interval of that payment, whether it was refunded, and the commission it earned them. They cannot see your business’s name, anyone’s email address, anything in your account, or any payment after the first.

Google Analytics runs on the public pages of this site — the home page, pricing, the free tools, the articles, sign-up and sign-in — to count visits and see which pages, tools and sign-up steps get used. It never runs inside the signed-in application, so nothing from your mailbox, your invoices or your customers can reach it; the one signal from inside is a completed subscription (plan, interval, amount and a Stripe reference), reported without the page’s content. Advertising features are off. Visitors in the European Economic Area, the UK and Switzerland are asked before any analytics cookie is set, everyone can decline, and a browser’s Global Privacy Control signal is treated as declining. Exactly what is sent, and the control to change your choice, are in the cookie policy.

We do not sell data and we do not share it with advertisers. Google Analytics is the only analytics product on the site, and it is not used for advertising.

Collectly AI is operated from the United States, and its providers — Neon, Netlify, Anthropic, Resend, Stripe — run infrastructure in several countries. Data reaching them may be processed outside the country you are in. If you need the specific regions in writing before connecting a mailbox, ask us and we will tell you what each provider account is actually set to rather than guess at it here.

• OAuth tokens encrypted at rest with AES-256-GCM, using a key held in the deployment environment and not in the database.
• Passwords stored only as bcrypt hashes. Password reset and email verification links are stored only as SHA-256 digests, so a leaked database yields no working links.
• Session cookies are httpOnly and unreadable from page scripts.
• Every request re-verifies which workspace you belong to on the server, so data cannot cross between organisations. There is a test suite specifically covering that isolation.
• Changing your password invalidates every session issued before it, including on other devices.
• Transport is HTTPS throughout.
• Audit records deliberately exclude message bodies, amounts and tokens.

Collectly AI holds no security certification — no SOC 2, no ISO 27001 — and does not claim one. The security page says what is and is not true in more detail. No system is perfectly secure. If we become aware of a breach affecting your data we will tell you promptly and describe what happened.

Disconnecting Gmail in Settings does two things: it asks Google to revoke the grant, and it clears the stored tokens. The local clear happens either way, so a disconnected mailbox never keeps a usable credential here. If Google does not confirm the revocation — a timeout, a network failure — the grant may still be live in your Google account, and the same applies to QuickBooks and Xero. When that happens Settings tells you so instead of showing a clean “disconnected”, and links the page that finishes the job. You can always remove it yourself from your Google account permissions page, and if you want certainty rather than the ordinary case, that is the place to check.

Both have a button in the product. Settings → Your data downloads everything tied to your own account as a JSON file; Settings → Danger zone deletes your account, or the whole workspace if you own it. The customer bundle is on each customer’s own page, for when one of your customers asks you. You can also email the address below and we will do it by hand.

• A copy of your data. Two bundles exist: everything tied to your own account, and everything one workspace holds about one of its customers.
• Deletion. Deleting a workspace deletes its invoices, customers, email threads and messages, analyses, follow-ups, timeline, invitations, provider connections, onboarding answers and feedback answers, and cancels its Stripe subscription. Deleting your account does the same for any workspace where you are the only member; where there are others, the workspace stays and ownership passes to a remaining member. Either way, your signed-in device records and your own feedback answers are deleted with your account. Your address is removed from the mailing list at the same time. Audit records survive with the account reference, IP and user-agent removed — an audit trail a subject can erase is not an audit trail.

Deletion does not reach the following, and we will not tell you otherwise:

• Backups. Neon keeps point-in-time history for its own window, and data deleted today stays recoverable from a backup until that window passes.
• The Stripe customer record, kept on purpose: it carries paid invoices and receipts, which are tax records Stripe holds in its own right. Ask, and we will remove it by hand if your situation requires it.
• Anything already sent to Anthropic, Google, Stripe or Resend, each of which keeps its own records under its own terms.
• Netlify's request and function logs, which age out on Netlify's schedule and not ours.

Deletion revokes every external grant before it removes a row, and stops without deleting anything if it cannot — dropping the record of a grant while the grant is still live would be the worst of both. If that happens we tell you which grants are still live and where to remove them yourself.

Your own account name and email are on your account; ask us and we will change either. For one of your customers, the whole record is editable from that customer’s page: company name, email address, contact name, phone number, notes and the domains Collectly AI matches their mail on. An address linked to the wrong company can be unlinked from the same place.

Two of those are worth naming, because getting them wrong has consequences beyond a typo. The primary email is where follow-ups are sent and what inbound replies are matched against. A matching domain attributed to the wrong company pulls that company’s unrelated correspondence into the record. Both are corrected in place, and Collectly AI refuses an address that already belongs to a different customer.

What cannot be corrected is the content of an email somebody actually sent. Collectly AI stores what arrived; it is a record of a conversation, not a claim about a person, and editing it would make it worthless as either.

This is the harder request, because the person asking has no account here and the business holding their data has obligations of its own. Deleting a customer record would take their invoices with it, and invoices carry tax significance that a business is separately required to keep.

So there is a third option in the product, on the customer’s own page: redaction. It erases the personal data and leaves the financial record.

• Erased: their name, email addresses, contact name, phone number and your notes; every stored email body, subject and participant list; every AI summary and line of reasoning; the text of every follow-up drafted or sent to them; and the wording on their activity timeline, including their name inside it.
• Kept: every invoice number, amount, currency, issue date, due date and status change, with its date. The customer becomes a short non-identifying reference so those invoices still reconcile to one party.
• Redaction is irreversible. Collectly AI keeps no copy of what it erased, the record cannot be edited afterwards, any queued follow-up to that customer is cancelled, and nothing will analyse or contact them again.
• A record of the redaction itself survives in the security audit log — who did it, when, and how many rows it touched. It contains no name, address or message content.

The same limits as deletion apply and are not hidden: backups still hold the pre-redaction rows until their window passes, and anything already sent to Anthropic or Google is held by them under their own terms.

We may update this policy. The date at the top changes when we do, and material changes are communicated to account holders.

Questions about this policy, a copy of your data, or a deletion request: nikowaliaweimer@gmail.com.