Trust
Privacy and data handling
This page describes the actual implementation, not the aspiration. If the product changes, this page changes in the same commit.
What we store
- —Account email and a hashed password if you sign up. If you use the anonymous trial, we store only a random session identifier in a cookie.
- —The files you upload, the passages we derive from them, and the embedding vectors of those passages. Nothing else about the file — no thumbnails, no OCR of images.
- —Usage counters (answers this month, documents uploaded) so quotas can be enforced. We do not store full conversation transcripts for anonymous sessions.
How a document is processed
- —Conversion: your file is turned into Markdown, then split into passages of roughly 350 tokens with a small overlap.
- —Embedding: each passage is converted into a 1,024-dimension vector by a Workers AI embedding model. Vectors are stored in a Cloudflare Vectorize index, tagged with your namespace identifier.
- —Retrieval: when you ask a question, the question is embedded the same way, and the closest passages from your namespace are returned. The query is filtered by your identifier, so another account’s passages cannot be returned even if the model were asked to look.
- —Generation: your question plus those passages are sent to the language model provider. The full document is never sent.
Who touches your data
- —Cloudflare — hosting, database (D1), object storage of vectors (Vectorize), embeddings (Workers AI).
- —OpenRouter — routes each generation request to a model provider. Requests are stateless; we do not enable provider-side prompt logging where the option exists.
- —We do not run advertising trackers on article pages, and we do not sell data. There is no data-broker integration anywhere in the stack.
Training policy
- —Your documents are never added to a training set, by us or by our model providers under our configuration.
- —We do not use your questions to improve our own models. Aggregate, non-identifying counts (for example: how many documents failed to parse) are used for reliability work.
Retention and deletion
- —Deleting a document removes its passages and its vectors immediately; there is no soft-delete window.
- —Anonymous trial sessions expire with the cookie. Sign in if you want your brain to persist across devices.
- —To delete an account and everything attached to it, email us and we will complete it within 7 days, then confirm in writing.
Questions or deletion requests
Email hello@musepics.com. A human reads it.