Security
Someone on your team deletes the wrong document. Or a server fails overnight. Before you move your team's work into Hyperclay, you want to know what you would get back.
Your documents are stored on one server in Virginia. Every save you make in the browser is kept as a version. The whole server is backed up every night, and each backup is kept for 200 days. Private documents open only for the people you allow and for Hyperclay administrators. Files you upload are public to anyone with their address. Four features use two AI providers, and each feature runs only when someone starts it. Add CMS sends a document to DeepInfra to add an editing panel to it. AI edits send your instruction and part or all of a page to DeepInfra to edit the open page. Page tools send your request and a short outline of the page to DeepInfra to make a small tool on the page. Describe anything sends the description you write, your answers and any short comments to Anthropic to build a new document. When generated follow-up questions are on, it first sends your description to DeepInfra to write those questions.
Where your data lives
Hyperclay runs on one Hetzner Cloud server in Ashburn, Virginia, USA. Your documents, uploads and account data are stored there. Pages on a custom domain pass through a second Hyperclay server, which forwards each request to the server in Virginia and keeps no copy of your documents.
Every plain HTTP request to hyperclay.com or one of its subdomains is redirected to HTTPS. The certificates come from Let's Encrypt.
Versions and Trash
Every save you make in the browser keeps a version of the document. The document's Versions panel lists the latest 50, and you can restore any of them. Versions are kept until the document is permanently deleted. Saves from the Hyperclay Local desktop app, and from tools that write through Hyperclay's sync API, do not create versions.
When you delete a document, it goes to Trash, in the account menu. The document's owner or a team editor can restore it from there for 30 days. After 30 days in Trash, a document and its versions are permanently deleted. When a team is deleted, all of its documents, including those in Trash, are permanently deleted 30 days later, with their versions.
Emptying Trash deletes the document and its saved versions from the live server. Copies remain in backups until those backups expire.
Backups
Every night at 04:00 UTC, the server copies the database, every document, every upload and its own configuration to Amazon S3. Amazon S3 encrypts the backups it stores. The server also takes a backup before every server update. Each backup is kept for 200 days.
If the server is lost, it is rebuilt from the latest backup. Changes made after that backup are lost.
Accounts and sign-in
Passwords are stored as bcrypt hashes. Hyperclay never stores the password itself.
When you sign in, the server sets a cookie that lasts 30 days. The cookie is marked secure, so it only travels over HTTPS. It is also marked httpOnly, so scripts on a page cannot read it. Each cookie matches a token stored on the server, and the server can revoke that token to end the session.
When a request that changes data carries your sign-in cookie, the server checks which site sent it and refuses requests from other websites. Form submissions to a document and Hyperclay's newsletter sign-up skip this check, because public forms send them from other sites.
Who can open a document
A private document opens only for:
- its owner
- members of the team that owns it
- people it is shared with by name
- anyone holding a share link to it
- Hyperclay administrators, the people who run Hyperclay
Share links are off until someone turns one on. Only the document's owner or a team editor can. A share link stops working once it is revoked.
Files you upload are not private. Anyone with a file's address can open it, even when the document that uses it is private.
AI
The document's owner or a team editor can start Add CMS, when the workspace is on a paid plan. When someone starts it, Hyperclay sends three things to DeepInfra: the document's URL, its file name and a copy of its HTML. The model is DeepSeek V4.1 Flash, and DeepInfra runs it on its servers.
A signed-in person can use Describe anything to have a new document built from a description. Generated follow-up questions are optional. When they are on, Hyperclay sends the description to DeepInfra after the person writes it. DeepInfra runs DeepSeek V4.1 Flash, an open-weights model hosted in the US. The model writes a few follow-up questions about what the description leaves open, each with a suggested answer. The person answers them or keeps the suggestions. Writing the questions is free to the person. If the questions cannot be written, the person gets the standard guided questions instead, and nothing more is sent to DeepInfra. Hyperclay sends their brief to Anthropic: the description they wrote, their answers to the guided or follow-up questions and any short comments they added. The model is Claude Opus 5.5, and it builds the document. Hyperclay then checks the built document in a sandboxed browser on its own server before saving it. Under Anthropic's commercial terms, what Hyperclay sends through its API is not used to train Anthropic's models.
The document's owner or a team editor can start AI edit or page tools, when the workspace is on a paid plan. For AI edit, someone presses Cmd+J in edit mode on a hosted document and describes a change. Hyperclay always sends DeepInfra the instruction they typed and the HTML of the element they are editing. On top of that, it sends any text they selected inside that element. If they reference the whole page with @page, and only then, it also sends the page's full HTML up to 400 KB. The edited HTML comes back and is applied to the open page. For page tools, someone asks a document to make itself a small tool, such as a calculator or a filter for a table. Hyperclay sends DeepInfra their request, a short outline of the page and, when they revise a tool, that tool's current code. The outline covers the page's title, its headings and the shape of its tables, capped at 16 KB. The model for both is DeepSeek V4.1 Flash, an open-weights model hosted in the US, and DeepInfra runs it on its servers.
Companies that handle your data
- Hetzner hosts the server.
- Amazon S3 stores the backups.
- Stripe processes payments. Card numbers go to Stripe and never touch Hyperclay. Hyperclay stores only your Stripe customer and subscription IDs.
- Postmark delivers the email Hyperclay sends.
- Google Workspace receives email sent to hyperclay.com addresses, including security@hyperclay.com.
- DeepInfra runs Add CMS, AI edit and page tools when someone starts them. When generated follow-up questions are on in Describe anything, it writes those questions.
- Anthropic builds the document in Describe anything when someone starts it.
- Cloudflare serves Hyperclay's marketing pages and the ClayJS scripts that make documents editable.
- Google Analytics loads on some of Hyperclay's own pages, such as the Privacy Policy and Terms. It does not load in the dashboard or in your documents.
Getting your documents out
To download one document, open its actions menu in the file list and choose Download. Inside a document, the document menu has the same item.
To download a whole workspace, open the account menu and choose Export workspace. Anyone who can see the workspace can export it. You get one zip with every current document, upload and folder in it. The zip does not include saved versions or Trash.
Not yet
Hyperclay does not offer single sign-on, audit logs or SOC 2 yet. If your team needs any of these, email security@hyperclay.com and tell us what you need.
Contact
Report a security problem to security@hyperclay.com.