A concierge, in your house style, drawing only from your archive.
Masthead gives luxury publishers an on-site assistant that composes citable dossiers from the publication's own writing — built, operated, and calibrated for each tenant.
A dossier, composed from a thirty-article archive.
A live instance, composing against a demo archive of thirty-odd articles across travel, watches, fashion, design, and hospitality. Each dossier is generated in real time. Every claim you read below is drawn from a specific article we've indexed.
The demo composes against a deliberately cross-vertical archive so any reasonable query returns a result. A production tenant composes against its own archive only; depth replaces breadth.
02 — What Masthead is
01
Composed from your archive
Masthead draws only from articles you've published. Every sentence in every dossier is citable back to a specific piece in your catalogue. There is no open-web retrieval, no hallucinated restaurant, no invented watchmaker. If it wasn't in your archive, it isn't in the dossier.
02
Your voice, calibrated
Before launch, your editor reviews twenty representative compositions and we tune prompts, vocabulary, and structural preferences against the notes. Calibration continues monthly thereafter. A reader cannot tell a Masthead dossier apart from a short piece written by your staff — and that's the point.
03
Your commerce, your affiliates
Every dossier closes with commerce cards — restaurants, hotels, watches, pieces, tickets — routed through your existing affiliate partners. Revenue flows to you, not through us. We earn against the build and against revenue share or a flat fee, at your choice. No take-rate at the click layer.
03 — How it works
For the CTO in the room.
Architecture
Each tenant is a first-class instance with its own database schema isolation (Postgres row-level security), its own theme, its own prompt calibration, and its own commerce tables. Masthead is a multi-tenant platform in the operational sense; your readers see only your brand, and your archive is never retrieved by another tenant's queries.
Models and cost
Composition runs on Anthropic's most capable model (currently Claude Opus 4.7). Classification and reranking run on a faster, smaller model. Every dossier persists its token count and resolved cost so per-tenant economics are transparent. Typical dossiers compose in three to six seconds and cost low single-digit cents at list Anthropic pricing.
Deployment
You point a subdomain — typically concierge.yourpublication.com — at our infrastructure via CNAME. We deploy on Vercel in the region closest to your readers, ingest your archive from a WordPress export (other CMSes on request), and hand you an admin URL for editorial review. No engineering work on your side beyond DNS.
Ingestion and re-ingestion
We import your archive once at onboarding and schedule a weekly re-sync thereafter. New articles appear in retrieval within twenty-four hours of publication. Embeddings re-run on any article you edit. Corrections propagate; dossiers cite current canonical text.
04 — The operational split
What Masthead does
Build the tenant: theme, prompt calibration, commerce routing, deploy.
Operate the instance: uptime, monitoring, cost control, model upgrades.
Re-ingest the archive weekly; resolve takedowns and corrections within a business day.
Review composition quality monthly and surface drift before it shows in dossiers.
Report tokens, cost, dossier volume, and commerce click-throughs to your team.
What you do
Provide the archive once at onboarding. Publish normally thereafter.
Own the affiliate partnerships. Provide us the link templates; we route.
Your editor reviews twenty dossiers during calibration week.
Your commercial team approves commerce card additions.
You keep the revenue from commerce click-throughs.
05 — What it costs
One-time build fee — $25,000.
Revenue share — 33% of attributable commerce revenue.
Flat monthly fee — available in lieu of revenue share. Contact for rate.
Build fee covers architecture, prompt calibration, archive ingestion, theme, commerce integration, and six weeks of post-launch tuning. Pricing is shown as a starting point; specifics negotiated per tenant.