Your data, your experts, your record.
How Eben Green keeps each workspace private, as it works today.
What we collect and why is in the Privacy notice.
Separate workspaces
Every database query is confined to its own workspace by row-level security, so one workspace can never read another's data.
Invite only
Testers, reviewers and admins join with single-use links. Each role sees only what it should, checked on every page, API call and data query.
A record of every decision
Who rated, approved, rejected or corrected each reply is kept with it. Every pull through the API is logged with its time and address.
Keys stay protected
Model and judge keys are encrypted at rest. Passwords and API keys are stored only as hashes, and a new API key is shown once.
Your models, your keys
Eben Green calls the model and judge you set, with your own keys. Model addresses can't point at internal networks.
What AI sees
Suggested tasks and suites are written from your project description alone. If you ask AI to read a free-form eval document, that document goes to our AI provider for that request.
Where it runs
The app runs on Google Cloud in Northern Virginia, with a managed Postgres database. Email goes through a transactional email provider.
Deleting data
Admins can delete a project and everything under it at any time: tasks, replies, reviews, reference answers, context and runs.
Have a security review to get through?
Book a call and we’ll go through your questions.