What private actually means here
Four separate claims, none of which follows from the others:
- Unguessable. The URL carries a 22-character random token. There is no sequential id to walk and no directory to enumerate.
- Unlisted. Nothing you publish appears in any public list, sitemap or feed — not even your own.
- Uncrawled. Every viewer response carries
X-Robots-Tag: noai, noimageai, noindex, and no referrer leaves with the reader. - Gated, if you want. A password, a work-email check, a single open, or an expiry — added in the publish call or afterwards.
What it is not
It is not end-to-end encryption by default. The bytes are stored so that a browser can render them, which means this deployment can read them. If you need the stronger property, publish with end-to-end encryption on: the key lives in the URL fragment, which browsers never send to a server, so the ciphertext is all that is stored.
The default, without you configuring anything
| Default | What it means |
|---|---|
| Visibility | Unlisted. The URL carries a 22-character random token; nothing is listed publicly. |
| Expiry | 7 days for a keyed publish, 24 hours for an anonymous one. Pass never to keep it. |
| After expiry | The row and the stored object are both deleted by a scheduled purge, not just hidden. |
| Robots | X-Robots-Tag: noai, noimageai, noindex on every viewer response. |
| Referrers | Referrer-Policy: no-referrer, so the token never leaks into someone else's logs. |
| Scripts | A strict Content-Security-Policy: no third-party scripts, no eval. |
| Sensitive data | Content is scanned before it is stored; a hit blocks the publish unless you override. |
| Transport | HSTS with a long max-age; the viewer is HTTPS-only. |
Why agents changed the requirement
When a person published a page, the page had been read by a person first. An agent publishes what it just wrote — a draft, a report with a client's numbers in it, a build that has not been reviewed. Public-by-default is the wrong setting for output nobody has looked at yet, which is why the private link is the plain path and not a paid feature.
Compared with the usual answers
| Approach | Where it leaks |
|---|---|
| A public static host | Indexed, enumerable, and permanent unless you remember to remove it. |
| A shared artifacts surface | Whoever opens it needs the vendor's account, or the page is public. |
| GitHub Pages | Private repositories still publish public pages; the gate is in an enterprise tier. |
| An email attachment | It lives in whatever mailbox it was forwarded to, forever, unrevocable. |
| This | Unguessable token, noindex, revocable, deleted on expiry. |
Publish something private right now
curl -X POST https://read.botook.ai/v1/publish -F file=@index.htmlNo account. The reply carries a link and a claim ticket; sign in later to keep the page.
With a key, from an agent
publish_html({ content: "<h1>Q3</h1>…", expires_in_hours: 168, password: "hunter2" })Stacking the gates
Expiry is deletion
A scheduled purge removes the record and the stored object together. A link that has expired
does not answer with a nicer page; it answers 410 and there is nothing behind it.
Revoking a link you already sent
Delete the site, or — if several audiences hold different links to the same page — revoke one share token and leave the rest working:
curl -X DELETE https://read.botook.ai/share-tokens/TOKEN_ID \
-H "Authorization: Bearer otr_live_YOUR_KEY"What the page itself is allowed to do
Agent-written HTML is code nobody read line by line, so the viewer runs it under a strict
content policy: no third-party scripts, no eval, no framing. Details on
the security page.
Sensitive content never reaches storage
Before anything is stored, the document is scanned for identifiers, card numbers, phone
numbers and dense lists of email addresses. A hit refuses the publish outright. Override
deliberately with pii_check: "warn" — the point is that the accident is caught, not
that you are prevented from publishing.
Who can see the page
- Anyone holding the link, subject to the gates you set.
- You, in the dashboard.
- Nobody else. There is no public directory and no shared workspace listing.
Analytics without surveillance
You get opens, unique viewers, scroll depth and whether readers reached the end. You do not get keystrokes, mouse paths or session replay, because none of that is recorded — what is and is not collected.
Where the link ends up without you noticing
Guessing the token is not the threat — it is 22 characters out of an alphabet of 62, which is past the point where brute force is a plan. The realistic way an unlisted page becomes a public one is that the link gets copied somewhere you did not intend. Some of that is ours to prevent and some of it is not, and the honest split looks like this:
| Where a URL leaks | Handled? |
|---|---|
| Into a third-party site's referrer log, when the page links outward | Yes — Referrer-Policy: no-referrer on every viewer response |
| Into a search index | Yes — noindex, and nothing links to it publicly |
| Into an AI training crawl | Yes — noai, noimageai, for whatever that is worth against a crawler that ignores it |
| Into a chat app's link preview | Partly — a gated page is never photographed, so the card is a placeholder |
| Into a forwarded email, a screenshot, a shared browser profile | No. Nothing technical can help here — that is what the gates below are for |
Which is the argument for not stopping at the unguessable link when the content matters: it is a good floor and a poor ceiling. A password, an email-domain gate or an expiry turns a leaked URL from a disclosure into a failed request.
Questions
Can someone guess my URL?
The token is 22 base62 characters. Guessing is not a strategy.
Does it need an account to open?
No. That is the point — the reader clicks and reads.
Can I use my own hostname?
Yes, with a CNAME and a TXT record: custom domains.
Is it really deleted?
Yes — record and object both. There is no archive to restore from.