Precedium

Privacy · September 2026

What Precedium stores, and what it never sees

This page says what data the workspace at app.deviceevidence.com holds, where it is held, for how long, and how to have it removed. It is written to be read. Where a statement is also enforced by the software, it says so. The terms are here.

  1. 1. Your documents never reach us. Test reports, protocols, drafts and anything else you add are parsed, indexed and stored in your own browser’s private storage on your own machine. There is no route on our server that accepts a file, and a build check fails if one is ever added. The site’s Content-Security-Policy header names the only addresses page script may contact: this site and the model providers listed in section 3, which a page contacts only when you draft on your own key with that provider. Your IT department can read that header without trusting us.
  2. 2. What our server holds about you. Two things, and only when you use the paid features. First, a session: your Consultant Licence key, in a cookie (pc_key) that page script cannot read, for thirty days or until you sign out. Second, if you buy a project: one row per purchase holding a hash of your key, the product tier, the token allowance and how much of it is used, and the Stripe payment reference; and one row per drafting call holding the token counts and the section name. Neither row contains your name, your email address, your documents or any drafted text. Your email address is verified against Device Evidence when you sign in and returned to your browser; we do not store it here.
  3. 3. Drafting with a language model. When you draft a section, the prompt is the sources shown on screen for that one section and nothing else. On the metered route it passes through our server to Anthropic’s API and is not stored by us. If you draft with your own key, the request goes from your browser directly to the provider you chose and does not pass through us at all; the site’s security policy allows page script to contact only the providers listed below, and a build check fails if the list and the policy ever differ. Your own key is kept in your browser only — for the tab session by default, or until you clear it if you ask the browser to keep it — and is never sent to us and never written into an archive.
    • Anthropic (api.anthropic.com) — Anthropic states that for API users it automatically deletes inputs and outputs on its backend within 30 days of receipt or generation, except where a longer-retention feature is used, a custom agreement applies, or usage-policy enforcement or legal compliance requires otherwise; a zero data retention agreement is available to eligible organisations. Inputs are not used to train Anthropic's models. Zero data retention: on request. Region: United States. Their terms, read by us on 2026-09-19.
    • Together AI (api.together.ai) — Together AI states that it does not use data collected from you to train its models without your explicit opt-in and consent, and offers a Zero Data Retention setting under Privacy & Security under which submitted content is not stored, retained or used for training and is removed as soon as processing concludes. Information may be transferred to and maintained on computers outside your country. Inputs are not used to train Together AI's models without explicit opt-in. Zero data retention: available. Region: United States (transfers outside your country possible). Their terms, read by us on 2026-09-19.
    • OpenRouter (openrouter.ai) — OpenRouter is an intermediary: your passages pass through OpenRouter and then to whichever host it routes the model to, so two data policies apply. OpenRouter states that it does not use your inputs or outputs for model training, that some model providers may, that its privacy page does not state a retention period for text prompts, and that data may be transferred to servers in the US or other countries outside the EEA. Its account settings let you exclude providers that may train on your data; set that before drafting. Inputs are not used by OpenRouter to train models; the upstream host's own policy applies and may differ. Zero data retention: no. Region: United States or other countries outside the EEA (regional endpoints exist for enterprise accounts). Their terms, read by us on 2026-09-19.
  4. 4. Archives. A .precedium archive is produced in your browser, encrypted with AES-256-GCM under a key derived from your passphrase by PBKDF2 with 600,000 iterations. We never receive an archive and cannot open one. A forgotten passphrase is an unrecoverable archive.
  5. 5. Analytics and cookies. The only cookies are the session cookie above and a theme preference. The hosting provider, Cloudflare, records request logs for security and operations and may attach a privacy-preserving analytics beacon that sets no cookie and stores no personal identifier. We use no advertising or cross-site tracking.
  6. 6. Retention and removal. The session cookie expires after thirty days. Purchase and usage rows are kept while the licence they belong to is live, because they are the record of what was bought and used. To have every row that belongs to your key deleted, or to ask what we hold, write to hello@precedium.com from the address on your licence; a person answers, and deletion is a real delete, not a flag.
  7. 7. Where the data is. Our server code runs on Cloudflare’s network; the two tables above are held in Cloudflare D1 in the European Union (Frankfurt). Your documents and projects are wherever your browser is.
  8. 8. Changes. If what the software stores ever changes, this page changes in the same release, and the month in its heading moves.