Privacy Policy
Last updated June 14, 2026
This policy explains what Prism collects when you run comparisons, how that data is used to improve the product, and the controls you have — including marking documents as Confidential to keep them out of any training data.
What we collect
We collect the prompts you submit and the model responses they generate in order to improve Prism. Each comparison run — your prompt, the selected models, and every model's full response — is recorded so we can evaluate quality and build better features.
Where it's stored
Comparison data is logged in our comparison_logs table in Supabase (Postgres). Documents you upload to a workspace are stored as sources, and their embedded passages in source_chunks to power grounded answers.
How your data may be used
Collected prompts and responses may be used to fine-tune our proprietary Judge model — the model Prism uses to analyze and score answers. We use this data to make the Judge and the overall product better over time.
Confidential documents
You can mark any document as Confidential from the composer's “+” menu before uploading it. Confidential documents — and any comparison run that draws on them — are excluded from training data. They remain fully usable for grounding your own answers; they are simply never included in any dataset used to improve our models.
Only you can see your data
Every table enforces Postgres Row Level Security (RLS). Policies tie each workspace, document, and comparison log to your account, so queries can only ever return your rows. Other users cannot read your workspaces, documents, or history.
Your controls
You can delete your documents and conversation history at any time. Deleting a workspace removes its documents and embedded chunks; deleting a conversation removes its comparison log. Deletions are permanent.
Questions
For privacy questions or data-deletion requests, contact the project owner. See our Terms of Service for the agreement governing your use of Prism.