Privacy Policy
Cairn stores things you want your AI tools to remember, so you stop explaining yourself over and over. This policy says exactly what it holds, who else can see it, and what you can remove.
It is written to be read. Where something is a genuine weakness, it is named rather than phrased around.
What Cairn holds
Things you chose to save. Every memory reaches Cairn as a proposal and is saved only after you keep it. Nothing is saved from a source without you reviewing it first.
The documents those memories came from. Cairn keeps the original text so that every answer can show the exact sentence it came from. These are encrypted before they are written to disk.
Your email address, to sign you in.
Records of activity — that a memory was approved, a source connected, an AI tool granted access. These hold no memory content.
Nothing else. Cairn has no advertising, no analytics of your content, and no tracking across other sites.
Who else receives it
Cairn is not a single company holding your data alone. Each service below receives a specific slice.
| Who | What they receive | Can they read your text? |
|---|---|---|
| Vercel (hosting) | Every request while it is being handled | Yes — this is where Cairn itself runs |
| Supabase (database, US East) | Encrypted memory and documents, plus metadata | No, except the items named below |
| Supabase Storage | Encrypted copies of original documents | No |
| WorkOS (sign-in) | Your email address and name | Identity only — never your memory |
| Google (Gmail, Calendar, Drive) | Only a request for what you connected | It is their data already; Cairn reads it |
| Sentry (errors, optional) | Error type and fixed diagnostic | No content |
| AI tools you connect | Only the memory they look up, with citations | Yes — see below |
If an AI model is turned on
This deployment runs with no external model at all: extraction, search, and answering all happen inside the application. No document text, memory text, or question you ask ever leaves this service to reach any AI provider.
If that ever changes, this section will be updated first, and the table above will name exactly who receives what.
AI tools you connect
When you connect Claude, Cursor, ChatGPT, or another tool, it can look things up in the memory you have saved. What it retrieves goes to that company, and from that moment their terms apply to it, not this policy. That is true of every AI tool, and it is why nothing is shared until you decide.
A connected tool can never change or delete your memory, and can never see anything still waiting for your review. You can turn any connection off at any time, and it stops working immediately.
What is not encrypted, and why
Your memory and documents are encrypted before they are stored. Some things cannot be, because a feature depends on reading them:
- Topic tags, types, and dates — filtering and sorting need them. These disclose subject matter and when you were active.
- Document titles and web addresses — a citation you cannot read is not a citation.
- Search embeddings — the numeric representations used to find relevant memory. These are stored unencrypted because the search index must compare them, and they are partially reversible: someone with direct access to the database could recover an approximation of your memory’s meaning from them. This is the weakest point in the design and it is named here deliberately rather than left out.
Google user data
Cairn’s use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.
Specifically:
- Gmail and Calendar data is read only to propose memories for you to review.
- It is never sold, and never transferred to anyone except as necessary to provide the feature you asked for.
- It is never used for advertising.
- It is not used to train any general-purpose AI model.
- No human reads it, except where you explicitly ask for support and grant access, or where the law requires it.
Disconnecting Google in Cairn destroys the stored credential immediately. You can also revoke access at myaccount.google.com/permissions.
How long it is kept
| What | Kept for |
|---|---|
| Memory you saved | Until you delete it |
| Original documents | 90 days, then the original is dropped — only the memory it produced remains |
| Activity records | 12 months, then removed |
| Sign-in sessions | 30 days, or until you sign out |
| Source credentials | Destroyed the moment you disconnect |
| Backups you download | Only by you — Cairn never keeps them |
What you can do
Export everything, as ordinary Markdown files, in one click. No request, no waiting.
Download an encrypted backup, locked with a passphrase only you know. Cairn cannot open it. If you lose the passphrase, nobody can.
Delete everything. Deleting your workspace removes memory, documents, sources, connections, connected AI tools, activity records, and finally the key that everything was encrypted with. You get a report naming each category and how many rows were removed.
Deletion is immediate. There is no grace period and no undo. If you want a copy first, download a backup or export before deleting.
Three things deletion honestly cannot reach, and Cairn says so rather than implying otherwise:
- Files you exported or downloaded yourself.
- Anything an AI tool already stored in its own history.
- The database provider’s own backups, for their retention window.
What Cairn does not claim
No independent security review
If that is not enough assurance for the data you were thinking of putting in, do not put it in yet. That is a reasonable decision and this policy would rather say so than talk you past it.
Changes, and getting in touch
Material changes will be shown in the app before they take effect, not published quietly.
Questions or deletion requests: johnmontejano2@gmail.com.