Privacy policy: what we store, and why

1. What this policy covers

This policy covers instibrain.ai, this marketing website, and the account plane at my.instibrain.ai that serves an account. instibrain is operated by WebWork AI, Inc., a Delaware corporation, which runs other products under their own policies.

Two different things are described below and they collect very different amounts. Reading this site collects essentially nothing. Holding an account collects what an account needs.

2. What this website collects

No cookies. No analytics. No trackers. No access log.

Cookies: this site sets none at all.
Server logs: our web layer is configured not to write an access log for instibrain.ai. Reading these pages does not put your address in a log file of ours.
Local storage: two keys, each only once you use it. theme records your choice on the light and dark toggle. guides records which lessons of the guides you finished. Both stay in your browser and never reach us. Clearing site data removes them.
Fonts: served from this origin. The two typefaces are downloaded from instibrain.ai and from nowhere else, so no font host learns that you visited.
Third-party requests: none. No content delivery network, no analytics tag, no embedded video, no social widget, no hosted font. The site's Content Security Policy starts at default-src 'none' and permits subresources only from this origin, so the browser would refuse a third-party request even if one were added by mistake.

The links that leave this origin are the Sign in button, which goes to my.instibrain.ai, and the links to our company's and our engineering partner's own sites in the footer and on the about page. Each is a link you choose to follow, not a request the page makes on its own.

3. What an account collects

There is no signup form. An account exists because one of us created it for a named person, by hand, after agreeing to do so, or because a person in an account invited you: an invitation is a one-time link we mail to you. The account plane stores:

Your email address, which identifies the account.
A password verifier, never the password. We store a scrypt hash with a random per-user salt, using the implementation that ships with Node.
A display name, and an avatar image if you upload one.
Your sessions. For each one: the time it was created, when it was last seen, when it expires, and the IP address and browser user-agent it came from. Those last two exist so you can look at your own session list and recognise which browser is which, and sign the others out. We store the SHA-256 of the session token, never the token, so a copy of our database cannot be replayed as a live session.
Password reset records, when a reset is requested: the SHA-256 of the reset token, when it was made, when it expires, and whether it was used. A reset link you ask for is valid for thirty minutes; the first link a new owner or an invited person gets is valid for seven days.
Two-step sign-in, if you turn it on: the authenticator secret, sealed under a key that lives only on our server; the hashes of your ten recovery codes; and a five-minute ticket while a sign-in waits for its code.
A pending change of email: the new address and a link valid for thirty minutes, until you confirm it from the new inbox. The old address gets a notice.
Invitations: the address invited, who invited it, the role and groups, and when the link expires. A person who never joins is removed after thirty days, or once every link has expired.
A record of each mail we sent you: the address, the subject, the time, and whether the relay took it, kept ninety days.
Rate-limit counters, keyed by your address or your IP address, kept up to a day.
The brain's report for the account: page counts and sizes, and for each computer you link its name, the time it linked, its last sync, and its sync client's version.
Activity counts: for each person, by the hour, the Claude Code sessions and prompts on their computers, in ~/instibrain and in other folders unless they decline it, their AIs' reads, the pages they change, and the share of their Claude plan's limits in use. A count holds no text and no path. The account's owners and admins read them.
Staff notes. Our staff may attach a note to an account on our internal cockpit, which only staff read.

There is no billing information, because nothing is sold. There is no waiting list, and no form on this site collects an address.

4. What your brain holds, and which spaces we can read

This is the part that matters most, and it is a choice you make for each space: open or sealed. Open and sealed spaces are described on the spaces page and the mechanism is on the security page.

Open space. We store the path and the body as markdown and our server reads them. It indexes the body so an AI can search it, and it keeps previous versions of a page. That is deliberate: it is what lets an AI holding no key reach a brain from a phone or a hosted surface.
Sealed space. Your computer encrypts the body and the path before they reach us, under a key we never receive. We hold ciphertext and cannot read it. What remains visible to us there is the name of each space and which space a page sits in, page identifiers, counts, approximate sizes, and timing. The limits section lists that residue rather than hiding it.

For an open space, each version of a page records which credential wrote it — an AI, a computer, or the cockpit — and the id of the person it belongs to, both as opaque identifiers. The space's managers, and the people who recover it, see them in its changes feed. A computer's sync writes carry its id in the page's version vector; a write through the cockpit or an AI carries the brain's own key there instead. For a sealed space the brain keeps no version history, and so no record of who wrote what. The brain's database holds no email address and no user-agent. It holds IP addresses in one place: the counters that rate-limit AI sign-ins, kept a day. The brain host's access log holds them too, and section 8 describes it.

Each time a page's text leaves the brain, for the cockpit, an AI, an agent, or a computer's sync, the brain records a page read: the page, who read it, through which credential, the kind of read, and the time. A search records how many pages its snippets showed, and a computer's pull names no page, so a sealed space's reads name none. The reads of a space are its read trail, kept thirty days. Its managers, and a personal space's person, see every row; anyone else the space is shared with sees their own; no AI, agent, or computer reads it. When one caller opens many pages of one space within ten minutes, the space's managers and the account's owners and admins get a mail. No read is stopped.

The brain reads the new text of every write to an open space once, in memory, against a fixed set of save rules: a secret of a known shape in any space; a card, IBAN, passport, or social security number; a server login outside its place; a line about an identifiable person's pay, assessment, or health in a space whose owner has not marked it as holding that; a link into a sealed space from a space others read. A write that breaks a rule is refused with the rule named. Nothing is written, nothing is kept, and the log line names the rule alone. When a space is shared, a group changes, or a person is invited, the brain reads each page a newcomer would read once more, so the person sharing can review it first. A specialist's pages are open pages, which the server reads. A proposal an AI leaves is stored as plain text until it is closed.

We do not use your content to train models. We do not sell it, rent it, or trade it, and we do not build advertising profiles.

5. How we use what we collect

To run the Service: authenticate you, keep a session, store and sync your content, and serve it to the AIs you connect.
To keep the Service safe: rate limiting, and investigating abuse or an incident.
To answer you when you write to one of our addresses.

That is the whole list. We do not profile you and we do not score you. The one automated decision is the save check in section 4: a rule refuses a write, the refusal names the rule, and you change the text and write again.

6. Cookies and browser storage

This website sets no cookies. The account plane sets two, and both are functional:

__Secure-ib_session: strictly necessary, set by our server. It holds a random session token and nothing else. HttpOnly so scripts cannot read it, Secure so it travels only over HTTPS, SameSite=Lax, path /, domain .instibrain.ai so the same sign-in works across our subdomains. It expires a week after you sign in, or after twelve hours unused, whichever comes first. Signing out deletes it.
sidebar_state: set by the page itself, remembering whether you collapsed the sidebar. It holds true or false, lasts seven days, and your browser sends it to my.instibrain.ai with each request; nothing on our side reads it.

The cockpit also keeps a few keys in your browser's local storage: brain.chosen and brain.held, the space you last opened and the spaces you hold, so it opens where you left off; brain.lastVisit, when you last opened it; notifications.seen, which notifications you have seen; and theme, light or dark. They stay in your browser, and signing out clears the brain. keys.

Session storage holds the token of a sign-in link while you finish the step it belongs to (instibrain.reset-token, instibrain.invite-token, instibrain.email-token) and a reload marker; both go when the tab closes. Our staff's own cockpit at management.instibrain.ai sets a cookie of its own, for staff; a customer never signs in there.

There are no analytics, advertising, or preference cookies to consent to, which is why you see no cookie banner. There is nothing a banner could meaningfully ask.

7. Third parties

One third party holds your data: the provider whose servers we rent holds the disks it sits on, and for a sealed space those disks hold ciphertext. No other third party processes it for us. Checked against our dependencies, not against intent:

No analytics or product telemetry of any kind, on either the website or the account plane.
No error or crash reporting service.
No payment processor, because nothing is sold.
No email delivery service. The account plane sends its mail through a relay we run on the same server: a new owner's first sign-in link, invitations, password reset links, the link that confirms a change of email and the notice to the old address, and a notice when your password changes, your other sessions are signed out, or two-step sign-in is turned on or off, and a notice when one caller opens many pages of a space within ten minutes. The relay signs each message with our domain's DKIM key and hands it straight to the recipient's mail server. Its log, kept like the other logs in section 9, records each recipient's address, and the account plane keeps the record of each mail named in section 3 for ninety days. Apart from mail, the account plane's only outbound calls go to the brain on the same machine: to review a share when a person is invited, to name the team spaces an invitation mentions, and when a person is removed or moved. Alerts about the server itself go to our own mailbox. One of them comes from a daily check of the brain's store, which reads ids, counters, and sizes, and never content.

No third party loads in the account plane. Its typeface and its icons ship with it, and its content security policy lets your browser fetch from my.instibrain.ai alone. The marketing site you are reading is bound by the same rule.

8. Data security

In transit: HTTPS everywhere, with HSTS. Plain HTTP redirects.
Sealed-space content: encrypted on your computer with ChaCha20-Poly1305 under a key derived per page. Each sealed space's key is wrapped with X25519 and ML-KEM-768 together, so a recording made today does not become readable when one of the two falls. The key never reaches us.
Open-space content: encrypted in transit and protected by access control, but readable by our server by design. Do not put something in an open space that you would not want us to be able to read.
Passwords: scrypt hashes with per-user salts, compared in constant time.
Session and reset tokens: stored as SHA-256 digests, never in the clear.
Application logs: our logger accepts a fixed set of fields and has no free-form escape hatch. It records an event name, a request id, the matched route pattern rather than the URL, a status, timing, and opaque identifiers including your user id. It cannot record an email address, a token, a password, a page path, or your content. It records an IP address in one case: a failed sign-in or a failed two-step code, so an attack on an account is visible. Database driver messages are reduced to an error code first, because a driver message can carry the value that caused the error.
The brain host writes an access log. Requests to brain.instibrain.ai are logged with the caller's IP address, user-agent, method, status, and the request path. The query, which names pages and search words, and the address a redirect points to are cut from each line before it is written; the other request headers stay. Beyond that log, an IP address sits in the rate-limit counters of both planes, kept up to a day, and in the application log on a failed sign-in. We would rather name each place than have you find it.
Network boundaries: our staff's cockpit sits behind a list of known addresses and a staff sign-in of its own. The account plane's path to the brain is reachable only inside the server's private network; the edge answers 404 for it.

No system is perfectly secure, and we make no claim that ours is. The security page states what the design covers and what it does not, including which requests are signed: a computer's, with its device key; an AI's carries a bearer credential instead.

9. Retention

Stated as it actually works, which means some of it is manual:

Sessions: expire a week after sign-in, or after twelve hours unused, and expired rows are swept automatically.
Password resets: a reset link expires after thirty minutes, a new owner's or an invited person's first link after seven days; spent rows are swept.
OAuth grants: authorisation codes, expired access tokens, and clients registered but never used are swept on a timer.
Your content: kept until you delete it. Two caveats worth knowing. A page in an open space keeps its previous versions, whichever way it was written: every version for 30 days, then one a day for a year, then one a week. So replacing its text does not remove what it said before; the space's content owner can erase the old versions under a path at once, a removed page's included. And deleting a page writes a tombstone rather than removing the row, so the fact that a page existed outlives its content.
The read trail: each page read is kept thirty days, then erased. Erasing a page's old versions erases its reads with them.
Account data: kept while the account exists. Four things age out on their own: a sign-in left in no account is removed within the hour; a person who never joined is removed after thirty days; the record of mails sent is kept ninety days; our staff's trail of what they did on an account is kept a year. There is no delete button for the account itself. Ask us at privacy@webwork.ai and we will erase the account from both planes. The mail record and the staff trail outlive it for their term. Removing a person takes their sessions, reset records, computers, AIs, and grants with them at once.
Backups: the databases are dumped nightly on the same server, and the most recent fourteen are kept. A copy of who may reach what is also kept each minute, for a day, so a restore keeps every access withdrawn since the dump. Each dump and each copy is encrypted as it is written, to a key that is not on the server. No third party holds them. A deletion therefore takes up to about two weeks to age out of backups, and we will not restore a deleted account from one.
Logs: application logs and the brain access log are kept by the container runtime on our server, at most five files of 10 MB per service, the oldest overwritten first. We have set no retention period in days, so how long a line lasts depends on traffic.
Email you send us: kept as long as needed to deal with it.

10. Your rights

Depending on where you live, you may have rights under the GDPR, the UK GDPR, or the CCPA and CPRA: to know what we hold, to get a copy, to correct it, to delete it, to restrict or object to processing, and to be free from discrimination for exercising any of them. We have no personal information to sell, so there is nothing to opt out of selling.

Write to privacy@webwork.ai and we will act on it. We aim to reply within thirty days. We may need to confirm you control the account before we act on a request about it.

One practical limit, stated because it is real: for the content of a sealed space we hold ciphertext and no key. We can delete it, and we cannot produce it in readable form for an access request. The readable copy is the one on your computer.

11. International transfers

Our servers are in Europe and the company is in the United States, so data is accessed from both. We have signed no Standard Contractual Clauses or other transfer agreement. If you need one before you use the Service, write to privacy@webwork.ai.

12. Children

The Service is for organisations and is not directed at children. We do not knowingly collect personal information from anyone under 18. If you believe a child has given us information, write to privacy@webwork.ai and we will delete it.

13. Do Not Track

Some browsers send a Do Not Track signal. There is no agreed standard for responding to it, so we do not act on it specifically. It makes no practical difference here: this site runs no tracking to switch off.

14. Changes to this policy

We will update this page when what we do changes, and change the effective date at the top. For a material change affecting account holders we will tell them directly. The rule we hold ourselves to is that this page changes before the practice does, not after.

15. Contact

WebWork AI, Inc.
1207 Delaware Avenue, Suite 513
Wilmington, DE 19806, United States

Privacy: privacy@webwork.ai
Security: security@webwork.ai
Legal: legal@webwork.ai
Anything else: hello@instibrain.ai

What we collect, what we store, and what we cannot read.