repryntt · ai158z
The repryntt agent
What our agent does on the web, how to recognise it, and how to reach us.
What it is
repryntt is its owner's own AI that acts, on their own computer, with receipts. When its owner gives it a task, like comparing prices, filling a cart or looking up opening hours, it may open a web browser and do that task the way a person would. It acts for one owner at a time, on tasks that owner asked for. It does not crawl, index or harvest sites.
How to recognise it
Page loads and data requests (document, XHR and fetch, over HTTPS) from the repryntt agent's own browser are signed with Web Bot Auth (HTTP Message Signatures, RFC 9421, Ed25519):
Signature-Agent: "https://repryntt.com"- Public keys:
https://repryntt.com/.well-known/http-message-signatures-directory - Signatures cover the host and the Signature-Agent header, and expire after one minute.
- Images, stylesheets and other files are not signed.
The signing key is held on repryntt's own servers, never on anyone's computer. A repryntt that signs asks our servers for a signature for each site it visits, and our servers apply rate limits and can stop signing for an account, or for a site, at once. Signing is off by default on a repryntt someone installs, and it is available only on paid plans, so every signed request comes from a paying account we can stop.
When our servers will not sign for a site, because it is on our blocklist or signing is switched off for that account, the agent's browser does not visit it at all. It never sends that traffic unsigned instead, and once the browser has been refused, the agent's web fetches skip that site too. Only if our servers cannot be reached for a moment does it carry on as an ordinary browser, without a signature. Our own test runs of the agent are not signed.
The browser also identifies itself as an automated browser. It does not hide the automation flag or claim to be a regular desktop browser.
How it behaves
- It takes no for an answer. If a site refuses it (401, 403, 429 or 503), it does not retry that page or try to get around the refusal. It tells its owner the site said no.
- It never solves challenges. CAPTCHAs and “are you human” checks are handed to its owner, a person, or the task stops.
- It never enters a card number. Card numbers are typed by the owner. It enters a password or one-time code only for its owner’s own accounts: from a login the owner saved and allowed it to use, or a code sent to the owner’s own email.
- It asks before it commits. Purchases, sending messages, deleting things and changing account settings wait for the owner’s approval.
- It does not post where people talk to people. On Hacker News, Reddit and lobste.rs it reads, but never posts, comments or replies.
- It browses at a person’s pace, one page at a time, not in parallel floods.
- It reads robots.txt when it fetches on its own. Background fetching follows robots.txt under the user-agent token “repryntt”. The browser that acts on its owner’s request is a user agent, not a crawler: it opens the pages that one task needs, and stops when a site refuses it.
- It reports what it did, keeps a signed receipt of its actions for its owner, and cites only pages it actually loaded.
Whose browser it uses
Its own, on its owner's device. When a task needs one of the owner's accounts, the owner signs in there once and it stays signed in. The same rules apply: no challenge solving, no card numbers, passwords and codes only for the owner's own accounts, and approval before anything that commits. The repryntt Link extension, which once let it use a tab in the owner's own Chrome, is off.
Blocking or limiting it
A site can check the agent's signature against the key directory above, and allow, rate-limit or block it by that signature. A site can also ask us, at the address below, to put it on our blocklist, and then the agent's browser does not visit it. We are applying to Cloudflare's verified bots and agents program; until Cloudflare approves it, Cloudflare does not list it as verified. When it's refused, it stops.
Contact
Questions, problems or abuse reports: hello@ai158z.com. We answer within 2 business days, and we can stop signing for an account the same day.
Operated by ai158z. Last updated 16 October 2026.