How Webtrax uses AI
Draft. Webtrax's AI features are not live yet. This page describes how they will work, and it will be updated and confirmed before the private beta.
What reaches a model
Excerpts of the public pages we crawl for you, and the findings about them (for example "this link returns 404"). For fix proposals, the relevant parts of your repository that you have connected. Nothing else.
Redaction
Before anything is sent, we remove personal data (email addresses, phone numbers) and anything that looks like a secret (API keys, tokens, passwords).
Which provider
With your own key, data goes only to the provider you chose. With Webtrax AI credits, we use providers that contractually do not train on API data, in the EU region where available. You can turn AI off for any site.
No training
We never use your sites, findings or code to train models, ours or anyone else's.
You approve changes
The agent proposes; you decide. A fix pull request changes nothing until you merge it. Direct changes need your approval, unless you switch on auto-fix for specific low-risk actions, which has daily limits and an undo.
Evidence and logging
Every claim cites a finding, a URL or a metric. Any URL the agent suggests is checked by our crawler before you see it. Every agent step is logged: what it saw, what it proposed and what changed.
Questions
Email hello@webtrax.dev.