Skip to content

Security and trust

Security in Minwal is part of the design, not a layer on top

This is what is built and tested in Minwal today. We claim no certification we do not hold, and we are glad to walk your security team through our security review in detail.

  1. The server decides

    Every authorisation is settled by the API; the screens only hide what is unavailable. A test walks every route to prove its guard is in place.

  2. History is never rewritten

    Published versions, a request's history and audit events are never edited.

  3. It fails closed

    With no organisation established, no data is read; when a file cannot be scanned, it is refused rather than kept unchecked.

Organisations kept apart

  • Every table an organisation owns carries its id, and its scope is applied automatically; a test fails the build if a table is added without it.
  • The organisation comes from the host name it is reached on, never from a value the browser sends.
  • Another organisation's records do not exist for you, and this is proved on every route.

Sign-in and sessions

  • Accounts lock for 15 minutes after five wrong passwords, with limits per network address too.
  • Usernames and mobile numbers sign in under the same limits, lockout and single answer as email, and a password an administrator types follows the one policy: 12 characters, upper and lower case and a digit.
  • Two-step verification with an authenticator app; each code is accepted once.
  • Single sign-on with OpenID Connect, the authorization code flow with PKCE, and the token's signature, audience and times verified. Signing in never creates an account.
  • The web session is an encrypted HttpOnly cookie, and changing a password signs out every other device. Phone tokens rotate on every use, and reusing an old one ends the whole sign-in.
  • Invitation and reset links work once, are stored hashed, expire, and never reveal whether an account exists.

Files

  • An allowlist of file kinds checked by their content, not only their name, and never SVG or HTML.
  • A ClamAV scan of every file before it is kept; when the scanner cannot answer, the upload is refused.
  • Stored files are scanned again every night with newer signatures, and one found harmful is quarantined.
  • Files are private and served only through their request, to the people who may see them.

Calls to other systems

  • Webhooks signed with HMAC-SHA256 over the time and the body, never carrying sensitive fields or comments.
  • Outbound calls go over HTTPS to public addresses only, checked at the moment of the call, with no redirects followed, a time limit and a limit on the size of the answer.
  • Directory sync over LDAPS or StartTLS only, with the server's certificate checked.
  • API keys with scopes, which never accept a browser session.

AI, when you switch it on

  • Off by default; nothing is sent until the organisation's administrator switches it on. On your own servers it can be taken out of the installation, so nothing is ever sent.
  • Only what a feature needs goes to the provider you choose: for a draft, the service's description and the names and codes of your reference data; for a summary, the service's name, the request's state, its stage names, the fields its reader sees and the kinds of events in its history.
  • Never a person's name, number or contact details, a key or a password. Confidential requests are left out unless you change that.
  • The text of requests and descriptions is treated as data, never as instructions; what a model returns is checked by Minwal's own rules or kept as plain text, and nothing of it is run or published without a person.
  • Keys are encrypted and never read back once saved, and never enter logs or error messages. Every call is in the audit log without its content, and each organisation has a daily limit.

Audit and data

  • A hash-chained audit log per organisation that never stores passwords, codes or secrets, with phone numbers masked.
  • A printout is built from its reader's own view, and its public check shows no name and no data.
  • Report exports write every text as text, so nothing typed can run as a formula in Excel.
  • Retention policies and legal holds, off until you switch them on.

The web app and the server

  • A Content Security Policy with a fresh nonce for every page, no framing, and no tokens or personal data in the browser's storage.
  • Containers run as unprivileged users; the database and Redis sit on an internal network only, and only the proxy faces outside.
  • Backups are readable by their owner only, and restored only after their checksums match.

Under load

  • Two people deciding the same task at the same moment never both count, proved with real processes on MySQL.
  • A load test with 500 people working at once, with no failed call.

What we do not claim

We hold no compliance certification, and no independent penetration test has been run yet. We share our security review and what we have verified with your security team, and we welcome your own testing.

See Minwal with services like your own

We will show you the product on services like yours, and answer the questions of your HR, IT and security teams.

Request a demo