Privacy

A manuscript in progress is unpublished text. This page says what Scribenda does with it, who it sends it to and how you get it back.

Your texts

Every book belongs to an account, and every read goes through a check tied to the current account. An automated test fails if an account's data can be read without that check, and another verifies that a user of one account reads nothing from another, neither from the site nor from the MCP server.

No book text goes into the application logs. Fields with book content are filtered out of request logs and background jobs receive only identifiers.

The model provider

Chat and semantic search go through OpenRouter, a service that forwards requests to the provider of the chosen model (today, OpenAI models). When you ask a question, both receive the question, the concept, the binding style guide and the book passages found by search. The service and model names are shown in the settings.

On every request the application asks OpenRouter to use only providers that do not collect user data.

Connected agents

An agent connected through MCP sees only the book and the areas you grant, and its model is your client's, not ours. Every action it takes stays in the book's log.

Uploaded files

Research files are recognized by their content, not their name, with a limit of 25 MB per file. They download through signed addresses that expire after a few minutes. Web pages are fetched with protections that stop requests from reaching internal addresses.

Download everything and delete the account

You can download each book as a zip archive with manuscript, bible, notes, sources and research in markdown, plus the original files. The download stays available on every plan, even after cancelling.

The account owner can delete it with all its data (users, books, research and files). The subscription is cancelled at the same moment.