Security at Schema

Last updated: November 9, 2026

A network diagram is a list of the things an attacker would most like to know. This page explains, in specifics, how Schema handles yours - including the parts we haven't built yet. If something here is unclear or you think it's wrong, write to admin@buildschema.app.

Schema never touches your infrastructure

Schema is a documentation tool. It has no agents, holds no credentials for your systems, and makes no network connections to your environment. It knows only what you put into a diagram. When you import an asset spreadsheet or an Nmap scan, the file is parsed in your browser; the file itself is never uploaded - only the diagram you choose to save is.

Encryption

  • In transit. All traffic between your browser and Schema uses TLS, and the site sends an HSTS header so browsers refuse to connect any other way.
  • At rest. The content of every cloud diagram - and every saved version of it - is encrypted with AES-256-GCM, using a fresh random nonce per write, before it's written to the database. The key is held as a secret on the application backend, separately from the database it protects, and can be rotated.
  • What that does and doesn't mean. This protects your diagrams if the database or a backup of it is exposed. It is not end-to-end encryption: the Schema backend decrypts a diagram in order to send it to you, so we are technically able to access content (for example to investigate abuse or answer a valid legal request). Diagram names are stored unencrypted so they can be listed.

Access control

  • Sign-in is handled by Clerk, a dedicated identity provider. Schema never sees or stores your password.
  • Every request for a diagram is authorised on the server against your verified session. The backend never accepts a user or organization id supplied by the browser.
  • A diagram you don't have access to is indistinguishable from one that doesn't exist - the response is identical, so nothing confirms that an id is valid.
  • Requests are rate limited per account, and there are caps on diagram size and count, so a script can't use your account (or the service) as bulk storage.

Audit trail

Creating, opening, changing, deleting or restoring a diagram, and creating, using or revoking a share link, are each recorded with who did it and when. A stretch of autosaves by one person is recorded as one change every few minutes, not one per save. Separately, every attempt that was blocked - an unauthenticated call, or a signed-in account reaching for a diagram that isn't theirs - is logged as a security event. These records are written by the backend as part of the operation itself; a client can't forge or suppress them. They're kept for up to 12 months.

Sharing

  • Diagrams are private by default. Nothing is shared until you create a link.
  • A share link is read-only, exposes only the view you choose (Manager, Engineer or both - the other view is never sent, not merely hidden), can be given an expiry, and can be turned off at any moment. Deleting the diagram kills its links.
  • Each link is a 256-bit random secret. We store only a SHA-256 hash of it, so a link can't be recovered from our database - which is also why the app can show it to you only once.
  • The secret sits in the part of the URL after the #, which browsers don't transmit to web servers, so it doesn't appear in hosting logs or referrer headers.
  • Anyone holding a link can open it, so treat one like a password and send it accordingly.

Version history and conflicts

Cloud diagrams keep up to 50 earlier versions, and any of them can be restored. If two people (or two tabs) edit the same diagram, the second save is stopped and you're asked what to do - Schema doesn't silently overwrite someone else's work.

Handling untrusted content

A diagram can come from a file someone sent you, a teammate, or a shared link, so Schema treats every diagram it opens as untrusted input. Files are validated and rebuilt field by field before they reach the canvas; links in a diagram are only ever clickable if they're plain http(s); exported reports escape all content and ship with a policy that forbids scripts; and CSV exports neutralise spreadsheet formulas.

The application itself

  • A strict Content-Security-Policy limits what the app can load and where it can send data, and the app can't be embedded in another site's frame (clickjacking protection).
  • No advertising or third-party tracking scripts. We count page visits with Vercel Web Analytics, which is served from our own domain, sets no cookies and never sees your diagrams or share-link secrets. The only cookies are the ones that keep you signed in. Fonts and icons are served from our own domain.
  • Dependencies are checked for known vulnerabilities, and backend access rules are covered by automated tests that run on every change.

Your data stays yours

From inside the app you can export any diagram (or all of them at once) in an open JSON format, and permanently delete everything stored in your account, without contacting us. You can also work without the cloud at all, saving diagrams as files on your own device. See the Privacy Policy for what we collect and how long we keep it.

Who else is involved

Schema runs on three service providers: Clerk (authentication), Convex (database and backend - where the encrypted diagrams live) and Vercel (hosting for the website and app). Each publishes its own security and compliance documentation.

What we don't have yet

We'd rather you hear this from us:

  • No third-party certification. Schema has not been through a SOC 2 or ISO 27001 audit, or an independent penetration test.
  • No end-to-end encryption, as described above.
  • No enterprise SSO (SAML) or team workspaces in the app today.
  • No contractual uptime or backup guarantee. Version history protects against mistakes; for anything critical, keep an exported copy as well.

If any of these is a blocker for you, tell us - it's how we decide what to build next.

Reporting a vulnerability

If you've found a security problem, please email admin@buildschema.app with enough detail to reproduce it. We'll acknowledge your report within three business days, keep you informed while we fix it, and won't take action against good-faith research that avoids accessing other people's data, degrading the service, or disclosing the issue before it's fixed. Our contact details are also published at /.well-known/security.txt.