Where a manuscript rests, and who can reach it.

Access is by seat on an imprint. A member of your staff holds a seat, and a seat reaches the titles on the imprints it covers — and nothing else. A title on another house’s imprint is not merely hidden from them; the request is refused server-side and answers identically to a request for a title that does not exist.

That last detail is deliberate. Telling a caller “this exists but is not yours” is a small disclosure, and a small disclosure repeated across a list of identifiers stops being small.

• Any title on an imprint the seat does not cover.
• An author’s private working conversations with the system.
• Any cost or token figure. Those are ours to carry, not yours to read.

AuthorsLab also operates a separate product for individual writers. The two share one engine and nothing else a user can see: a writer’s own work is protected row by row at the database, and a publisher’s access is resolved from their seat on every request and refused outside it. AuthorsLab’s own staff accounts can read across the platform, to support it.

A manuscript is stored in Supabase (database and storage) and read by the application on Vercel. The editorial reads run through n8n Cloud and are performed by Anthropic models. Reports are rendered by APITemplate, documents are converted by ConvertAPI, and email is sent through Resend and Titan; Cloudflare handles DNS and network protection. OpenAI draws cover images from a written description, not from the manuscript. Stripe processes payment and never receives manuscript content.

The full list, with what each one processes, is at /subprocessors.

We state a location only where we have checked it ourselves. Today that is Supabase and the application’s server code on Vercel, both in Singapore, and Resend’s sending, in Ireland. Where n8n and Anthropic process a manuscript we have not yet verified, and we say so rather than guess.

Decisions a publisher makes inside the system — an approval, a request for revisions, a note — are written to an append-only log, attributed to the named seat and the organisation that holds it. A publisher cannot record a decision under a name they supply; the actor is derived from the session, not from the request.

This exists so that “the house approved the cover” is a checkable statement months later, which is the only version of that sentence worth having.

A publisher-side privacy policy, terms and data processing agreement are being drafted. Until they are published, the operative documents are the ones linked in the footer, and anything in this page that they contradict should be read as this page being out of date.