Headers on every viewer response
content-security-policy: default-src 'self'; script-src 'self'; object-src 'none';
base-uri 'none'; frame-ancestors 'none'
x-robots-tag: noai, noimageai, noindex
referrer-policy: no-referrer
x-content-type-options: nosniff
strict-transport-security: max-age=31536000; includeSubDomainsVerify it yourself against any page you publish — it is one curl -I away.
Why the policy is strict by default
Agent-written HTML routinely contains a pasted analytics snippet or a CDN font nobody asked for. A strict policy means that a page which quietly reaches for a third-party script simply does not get to; the content still renders, the beacon does not fire.
The link itself is the first gate
A 22-character random token, never listed and never indexed. There is no id to enumerate and no directory to walk — what private means here.
Unguessable is a quantity, not an adjective: the token is drawn from 62 characters at 22
places, so the space is about 1039 wide. Guessing is not a threat model anyone has to
plan against; losing control of the link is, which is why the gates above exist and why every
page carries no-referrer so the token never leaks in an outbound request.
Gates you can add
| Gate | Checked | Reader needs |
|---|---|---|
| Password | Before any content is served | A passcode you gave them |
| Email domain | Before any content is served | An address at one work domain, confirmed once |
| Burn after read | On each successful open | To be first |
| Expiry | On every request | To arrive before the deadline |
Expiry means deletion
A scheduled purge removes the database row and the stored object together. Expired content is
not archived, not hidden, not recoverable — the link answers 410 and there is nothing
behind it.
The sensitive-data scan
Before a document is stored it is checked for identity numbers, card numbers, phone numbers
and dense lists of email addresses. The default mode blocks the publish; warn
publishes anyway and tells you; off skips the check. Blocking is the default because
the failure it prevents — a page nobody meant to send — is not recoverable once the link is out.
End-to-end encryption
When you need this deployment not to be able to read the content, publish encrypted. The key lives in the URL fragment, and browsers never send a fragment to a server, so only ciphertext is stored. The trade is real: lose the link and the content is unrecoverable, and no server-side feature — preview images, plain-text extraction, search — can work on bytes nobody can read.
Credentials
- API keys are stored as hashes and shown once. A lost key is replaced, never recovered.
- OAuth tokens are signed
HS256with a 30-day life. - Revocation kills the refresh token with the access token — a revoked grant cannot be traded back into a fresh one.
- Passwords on pages are hashed, never stored in plaintext.
Sessions in the dashboard
Browser sign-in is GitHub OAuth or a one-time emailed link. Session cookies are
HttpOnly, Secure and SameSite, and the browser session is
never accepted as API authentication.
Rate limiting
Anonymous publishing is limited per address in a fixed window, which is what keeps an open endpoint from becoming someone else's file host.
The infrastructure
Cloudflare Workers for the runtime, D1 for records, R2 for the stored documents, Queues and scheduled triggers for the purge and the digest. Nothing is on a third-party analytics or logging service — the parts of the system that could see your content are the parts that have to.
Where the content actually sits
The document goes to object storage; the metadata goes to the database. Both are in the Cloudflare account running this deployment. If that account is yours, then so is every byte of this — self-hosting is not a fallback here, it is the normal case.
Source-available, and self-hosted by construction
You can read the code that serves your pages, and you already are — this deployment is running from it. There is no privileged build with different behaviour.
What we cannot do for you
- We cannot stop a reader who legitimately opened a page from screenshotting it.
- We cannot un-send a link. Revoke it, expire it, or delete the page — that is the whole toolbox.
- Without end-to-end encryption, we cannot claim the host is unable to read the content, because it plainly can.
Check it rather than believe it
Every claim on this page is something you can measure from outside in about a minute, and a security page that cannot be checked is marketing. Publish a throwaway document and look at what comes back:
# the headers on a viewer response
curl -sD- -o /dev/null https://read.botook.ai/p/UNLISTED_TOKEN/ \
| grep -iE 'content-security-policy|x-robots-tag|referrer-policy|strict-transport|x-content-type'
# a gated page must not leak the body with the 401
curl -si https://read.botook.ai/p/GATED_TOKEN/ | head -1
# an expired page
curl -so /dev/null -w '%{http_code}\n' https://read.botook.ai/p/EXPIRED_TOKEN/What you should see: the strict policy plus the robots and referrer headers on every viewer
response; a 401 that carries the gate page and none of the document behind it; and for
an expired link a 410, becoming 404 once the scheduled purge has removed
the record and the stored object. The two codes are deliberately different — 410 tells
a late recipient the link really existed so they can ask for a new one, and a revoked share token
answers 404 instead precisely so it cannot be used to prove a page was ever there. If
any of that differs from what is written above, it is a bug and worth reporting.
Reporting something
Machine-readable contact at /.well-known/security.txt, or mail
hello@botook.ai. Include how to
reproduce; that is the part that saves everyone a week.
Privacy of the reader
Aggregate counters only: opens, unique viewers, scroll depth, whether anyone reached the end. No recording, no replay, no cross-site identifier — the full list.
Questions
Can you see my pages?
Whoever runs this deployment can read stored bytes, unless you published with end-to-end encryption. On a self-hosted install that is you.
Is content scanned for anything else?
Only the sensitive-data check described above, and only at publish time.
What happens on account deletion?
Records and objects go together; expiry and deletion use the same purge path.