Security and trust
Your research stays yours.
Your files and questions are never used to train a model. We have enterprise agreements for zero data retention with every model provider we use, including OpenAI, Anthropic and AWS. Access stays within the permissions you set.
At a glance
- Training on your work
- Never
- Kept by the models
- Nothing
- GDPR and DPDP
- We comply today
- SOC 2 Type II and ISO
- In progress
The data path
Where your work goes, and where it stops.
Your files, projects and research stay within your organisation. Access is controlled by your team, and sharing is deliberate: a link you create or an export you request.
the line around your company
Your files
Your projects
The research run
runs code on your data
A sealed workroomno internet access at all
a share link you created
Research you shared
an export you asked for
A file you exported
Never leaves: your data as training material
The promises
Two columns, and no small print.
Most terms of service hide the answers in small print. Here they are in two columns: what we do with your files, your research and the people who can open them, and what we never do.
What we do
- Keep your files in your own workspace.
- Read them only to answer your question.
- The models that read it keep nothing.
- Give each person only what their job allows.
- Work on your numbers in a sealed workroom.
What we never do
- Pool your files with anyone else.
- Train a model on your work.
- Keep hidden copies of your work.
- Open a file a person’s job does not cover.
- Send anything anywhere you did not point it.
Where we stand
We say in progress until the day it is not.
SOC 2 Type II and ISO are in progress. We are not certified yet, so we do not say we are. GDPR and DPDP are settled, and we comply with both today. The day a certificate is in our hands, this page will say so and carry the date.
Your security review
Your questionnaire
Answered in writingIn your formatBoth audits as they standNo marketing in between
Questions
Common questions
No. Your files, your questions and the research that comes back are never used to train a model. The models that read your work keep none of it once the answer is done. That holds on every plan.