Privacy and training data
What we collect
Once you sign in, each generation in the workbench records: your brief, the HTML the model generated, the background audit's scores, and your “Keep / Not good” feedback along with actions such as deploy, export and regenerate. Anonymous (signed-out) generations never enter training data.
Purpose
The only purpose is to improve the HTML generation model we train ourselves: a real distribution of requests and real failure cases are something synthetic data cannot provide. The data is never sold, and never handed to third-party model providers for training.
On by default, off at any time
Collection is on by default. You can turn it off in Settings. After that, new generations are no longer included, samples not yet frozen into a training set stop being exported immediately, and samples already in a training set are marked for deletion and removed at the next dataset rebuild.
De-identification and deletion
Before export, briefs are de-identified (emails, phone numbers, ID numbers, IP addresses and the like are replaced with placeholders); pages with too many hits are quarantined and not exported at all. User identifiers only appear as salted hashes. You can ask us to delete all of your data: contact us with the email you signed in with.
Observability data
For troubleshooting and billing, each request logs its trace id, request id, token counts, latency and status to our logging service (kept for 30 days). These logs do not contain your brief or the generated HTML.