Skip to content

Security

A BI tool sits in front of your company’s data, so Riff BI is built to read your warehouse and nothing else. This page explains how queries are kept read-only, how credentials and passwords are protected, who can see what, and what is recorded. What the AI features can see is covered in AI and your data.

AreaWhat Riff BI does
Warehouse accessRead-only. Every statement is checked before it runs, and anything that isn’t a read is refused.
Warehouse credentialsEncrypted before they are stored, and never shown again or sent back to the browser.
PasswordsStored only as salted scrypt hashes, never in plain text.
Connections in transitTLS to your warehouse, with certificate verification available for each connection.
WorkspacesEvery connection, model and dashboard belongs to one workspace, and every query is scoped to it.
Access controlPermission-based roles, checked on every request, including requests Rivo makes for someone. Four roles come built in, and custom roles can be set up on request.
Sign-inInvite-only, with throttling of repeated failed attempts and sessions that expire.
AuditAn audit log of who changed what, which Admins can filter and read.

Riff BI never inserts, changes or deletes data in your warehouse. This is enforced in more than one place, so a single mistake can’t turn into a write.

  • Every statement is checked before it runs. Only a single read is allowed. A statement that writes, changes the schema, changes permissions or changes session settings is refused, including one hidden inside a subquery or a WITH clause.
  • The warehouse enforces it too, where it can. On PostgreSQL, every session Riff BI opens is read-only at the database level, so a write would be rejected even if it got past the check.
  • Every query has limits. Queries are cut off after a time limit and return a capped number of rows, so one runaway query can’t tie up your warehouse.
  • Nobody writes raw SQL behind a chart. Charts are built from a data model’s fields, and Riff BI writes the query. Rivo and the dashboard assistants reach your warehouse only through the same checked, read-only, row-limited queries.

We still recommend connecting with a warehouse user that can only read the schemas Riff BI needs. See Connections.

A connection’s host is checked against the addresses it actually resolves to before Riff BI connects. Cloud metadata endpoints and link-local addresses are always refused, and private or internal addresses are refused unless the installation is set up to allow them. The checked address is the one Riff BI connects to, so a DNS answer that changes in between can’t send the connection somewhere else.

WhatHow it is protected
Warehouse passwords, keys and tokensEncrypted with Fernet (AES in CBC mode with an HMAC-SHA256 integrity check) before they are written to Riff BI’s database. The app shows which secrets are on file, never their values.
Account passwordsHashed with scrypt and a random salt for each password. Riff BI can check a password but can’t read it back.
Share-link passwords and tokensStored as hashes. Changing a link’s password ends every session opened with the old one.
Traffic to your warehouseTLS, set per connection. PostgreSQL and MySQL connections can verify the server’s certificate; Trino and ClickHouse use HTTPS by default, and Trino refuses a password or token over plain HTTP.
Sign-in cookiesMarked Secure, HttpOnly and SameSite=Lax, so they travel only over HTTPS and page scripts can’t read them.
  • Workspaces are kept apart. Every connection, model, dashboard and conversation belongs to one workspace, and every lookup is filtered by it. Asking for something from another workspace gets “not found”, so its existence isn’t revealed either.
  • Permission-based roles. Every action is a permission, checked on the server for every request. Admin, Modeler, Creator and Viewer come built in, and we can set up custom roles with the permissions your team needs (see Roles in Administration). Rivo acts with the permissions of the person using it, never more. See Roles.
  • Folder sharing. Folders decide which dashboards and models each person can see. See Sharing.
  • Lowering access takes effect at once. Demoting or deactivating someone signs them out of every session immediately, not at their next sign-in.
  • A workspace always keeps an Admin. The last Admin can’t be demoted or deactivated.
  • Invite-only. There is no public sign-up. People get in when an Admin invites them. Invitation links expire after 7 days, and password reset links after 24 hours.
  • Repeated failed sign-ins are throttled. After too many wrong passwords, further attempts are refused for a few minutes. The limit counts both the account and the address the attempts come from, so one client can’t work through a list of emails, and nobody can lock someone else out.
  • Sessions expire. A session ends after 14 days without activity, and after 90 days at most, however active it is.
  • Changing your password signs out your other sessions. An Admin can also use Sign out everywhere on any person in Administration.
  • Cross-site requests are refused. Requests that change something must come from Riff BI’s own pages, on top of the SameSite cookie setting.

A share link lets people without an account see a dashboard or folder, so it carries its own controls. See Sharing.

  • An optional password, with repeated wrong passwords throttled the same way as sign-in.
  • An optional expiry date, and the link can be turned off at any time.
  • Locked filters, so the person holding the link sees only the slice you chose.
  • A switch for Allow AI questions, with a limit on how many questions each visitor can ask. Those questions follow the rules in AI and your data.

Anyone holding a link and its password can see what it shows, so share links are best kept for data you’re comfortable sending outside your workspace.

  • Audit log. Sign-ins, role changes, deactivations, password changes, share-link changes, publishing and other changes are recorded with who made them, when, and from which address, including changes Rivo made for someone. It never holds passwords, secrets, SQL or warehouse values. Admins read it under Audit log.
  • Staff access. When Riff BI staff open a workspace’s assistant conversations to investigate a problem, it is recorded in a separate platform audit log.
  • Application logs. Riff BI’s logs never contain query text, query parameters or connection secrets.
  • Troubleshooting copies. Failed requests and AI runs are kept for 15 days so problems can be fixed. Any query saved with a failed request has its values removed, and successful queries are never stored.

By default, the AI model sees how your data is organised, not what is in your rows. Columns marked as personal data stay hidden even when an Admin lets the AI see sample values. Rivo proposes changes, and nothing is applied until a person approves it. The full detail, including which providers Riff BI may use and what each keeps, is in AI and your data.

If your company reviews vendors before using them, these features help:

  • A read-only warehouse user you control, so you decide exactly which schemas Riff BI can read.
  • The audit log, for evidence of who had access and what changed.
  • Personal-data marking on columns, which keeps those values away from the AI.
  • Per-connection control over whether the AI may see sample values, off by default.
  • Roles, folder sharing and share-link expiry, to keep access to the people who need it.

For a security questionnaire or more detail, write to us (below).

If you think you’ve found a security problem in Riff BI, write to support@riffbi.com with “Security” in the subject line. Please include enough detail for us to reproduce it, and give us a chance to fix it before sharing it publicly.