We change code on live websites. Here is why that is safe.
Clients grant us access to their platforms so we can fix accessibility at the source. That trust is the entire business, so the safety systems are not an afterthought. They are most of the engineering.
Six safeguards on every change
Every remediation job runs through the same pipeline, in the same order, with no exceptions and no manual bypass.
Sandbox before production
Every fix is applied to a test copy first and verified there. Nothing touches your live site until the change has already worked somewhere safe.
A snapshot before every change
Before any fix ships, the exact original content is stored in a snapshot. Not a description of the change. The literal HTML, byte for byte.
One-click rollback, any time
Any fix, or all of them, can be reverted from your dashboard. Rollback restores the snapshot exactly and records the reversal in the audit log.
Confidence gates on every AI fix
Each generated fix carries a confidence score. High-confidence pattern fixes you have already approved can apply automatically; anything below the review threshold stops and waits for human eyes. The thresholds are settings, not suggestions.
Design changes wait for you
If a fix would visibly change how your site looks, the system does not guess. It pauses the job, shows the before and after, and waits for your approval before anything ships.
An independent referee
After fixes apply, an independent third-party accessibility check (WAVE, by WebAIM) runs against the result. Any error it reports blocks completion until a human clears it. We do not grade our own homework.
How we hold your credentials
Platform credentials (WordPress application passwords, Shopify and Webflow API tokens) are encrypted at rest with AES-256-GCM using a key derived through scrypt. They are decrypted only inside the worker at the moment a job needs them, never rendered in a browser, never written to logs, and deleted when you disconnect an integration.
Where a platform offers scoped access, we use the narrowest scope that can apply a fix. On developer workflows, fixes arrive as a pull request that merges only after an approval: yours in the dashboard by default, or your team's code review if you enable required review. We never push straight to your default branch.
Every action a job takes, connect, scan, fix, verify, rollback, is written to an append-only audit log with timestamps, so there is always an answer to "what exactly did you change, and when?"
Infrastructure and data
- TLS everywhere, with HSTS preload. Content Security Policy, frame denial, and strict referrer policy ship on every response.
- Data lives in managed Postgres (Supabase) with row-level security policies; services authenticate with scoped keys.
- Payments are handled entirely by Stripe (PCI-DSS Level 1). Card numbers never touch our servers.
- AI fix generation uses the Anthropic API, which does not train models on API inputs. We send only what the fix needs: the failing snippet, not your database.
- Attribution uses a single first-party cookie. No third-party ad trackers, no session recording of your visitors.
- GDPR and CCPA rights (access, deletion, portability, opt-out) are honored through the dashboard and support@remediada.com, with a 30-day soft-delete window before permanent erasure.
The legal detail lives in our Privacy Policy and Terms of Service. To report a vulnerability or ask a security question, email support@remediada.com with "Security" in the subject line and we will respond within one business day.
Trust is earned in the audit log.
Run a free scan, read the findings, and see exactly how the fix pipeline works before you grant anything access.