Security and trust
Autoflowly builds apps that handle real customers and real money, and it lets AI agents act inside them. This page describes how accounts, data, apps and agent actions are protected, in enough detail to evaluate the platform and without publishing an internal blueprint.
Accounts and sessions#
- Sign in with Google, GitHub, or email and password. Passwords are hashed with a modern algorithm and never stored in clear text.
- OAuth exchanges happen server-side; client secrets never reach the browser or the phone.
- Sessions use short-lived access tokens with refresh tokens. Signing out invalidates the session on the device.
- High-risk approvals on the phone require Face ID, Touch ID or the device passcode before the decision is sent, and the decision records which was used.
Encryption#
- In transit: TLS everywhere, including between your apps and the platform. Plain HTTP redirects to HTTPS.
- At rest: credentials, provider tokens and connector secrets are encrypted with keys that live outside the database. Secrets you enter in the dashboard are write-only.
- Secrets never in code: generated apps receive credentials at runtime from the platform, never in their source. A generation request that tries to make the AI read secrets or reach the cluster is refused.
App isolation#
| Layer | How |
|---|---|
| Compute | Each app runs in its own container with resource limits; no shared processes between customers |
| Network | Default-deny policies; only the traffic an app needs is allowed |
| Data | A database per app, with a login role that cannot escalate privileges or see other databases |
| Address | A unique address per app, with a per-app or wildcard certificate |
| Drafts | A draft app answers only to its owner and to holders of a purpose-bound preview link; the preview link can never reach data or admin interfaces |
Governed agent actions#
Agents propose; the platform decides what may run.
- Risk floors are deterministic. Money actions are high risk whatever the agent says, and an amount above the threshold is too. Floors only ever raise risk.
- Three doors, one policy. Reading app data, proposing an action and executing an approved action each check the app's permissions. The execution door fails closed for money.
- Money mandate. A money action is automatic only under an explicit owner-set limit and below it.
- Daily cap on automatic actions; nothing automatic in sample mode or while the team is paused.
- Complete audit trail. Every proposal, decision and execution records the agent, the approver, the surface, the authentication used, the risk and a fingerprint of the exact parameters. Denials are recorded with the rule that denied them. Exportable per app.
- Growth money never moves without an approval event; a failed check pauses campaigns.
Pre-publish security check#
Before an app goes public the launch check probes it anonymously, the way a stranger would: is sign-up closed unless opened on purpose, are admin pages gated, are the legal pages present, does the draft stay private. Findings appear on the dashboard's Security tab, and a failing check blocks publishing until fixed.
Generated code safety#
Every build passes syntax, dependency, routing, build and render gates before deployment. A change that breaks a working app is never deployed; the previous version keeps serving. See Self-healing builds. Starters are rebuilt nightly from scratch to catch drift.
Platform hardening#
- Standard defensive headers on every response (strict transport security, no sniffing, frame restrictions, a restrictive content security policy).
- Schema validation on all API input; parameterised database access throughout.
- Rate limits by identity and tighter limits on sign-in and generation.
- Daily database backups with 30-day retention, and a snapshot before any schema change in production.
- Deploys are canaried: a new version serves a fraction of traffic under observation and is rolled back automatically on errors.
Compliance and data#
Autoflowly is operated from Germany and designed for GDPR. See Privacy and GDPR for what is stored, how to export or delete it, and what every generated app ships for its own visitors.
Private cloud#
Organisations that must keep apps in their own cloud account can run them on their own cluster with an outbound-only runner. See Private cloud and SSO.
Report a vulnerability#
Write to security@autoflowly.com. In scope: authentication and authorisation, data exposure across accounts or apps, injection, agent governance bypasses, infrastructure. Please give us a reasonable time to fix before disclosure; we acknowledge reports within two working days and keep you informed.