@plesso/content-factory

AI article drafter for editorial-grade content sites built on Sanity. Pairs with @plesso/sanity-seo — content-factory produces drafts; sanity-seo validates and tracks them.

Status

Phase 4.4 — sources lifted. All libraries, scripts, and Vercel-edge HTTP handlers are in src/. The package builds in isolation. Phase 4.3 (next session) switches surprisemywife to consume both @plesso/sanity-seo and @plesso/content-factory via published npm packages.

What ships here

Pure libraries (lib/)

• draftArticle(args) — the drafter. Calls Anthropic (model + prompt + temperature from args), parses the response into Markdown, converts to Portable Text, scores + inserts internal links via @plesso/sanity-seo/lib/suggestLinks, returns a complete draft object ready for commitDraft. Includes prompt template hydration, per-call telemetry capture, monthly budget caps.
• commitDraft(client, draft, opts) — writes the draft to Sanity as a draft document. Idempotent on the draft ID.
• isClusterDue(cluster) + isOverMonthlyBudget(cluster) — schedulers. Plain functions; no I/O.
• markdownToPortableText(md) + portableTextToPlain(pt) — bidirectional. The Markdown parser is intentionally small; assumes the model output stays within the prompt's stylistic bounds.
• notifySlack(payload, opts) — webhook ping for new drafts. Resilient: failures log and continue.

HTTP handlers (api/)

Framework-agnostic. Each handler takes a Web Request and returns a Response. Wire them into your runtime of choice.

• api/draft — single-draft endpoint. POST with cluster ID + topic. Returns the committed draft ID.
• api/scheduled-run — cron walker. GET (with Vercel x-vercel-cron header OR a shared secret) iterates every enabled cluster, drafts the ones whose cadence is due and budget allows. Aggregate report in response body.

CLI scripts (scripts/)

• factory-draft.mjs — local trigger for draftArticle. Loads .env.local, picks a cluster + queue item, drafts, prints the result. Use for one-off experiments without redeploying the edge endpoint.
• seed-content-factory.mjs — idempotently creates a default promptTemplate and a starter topicalCluster (both schemas live in @plesso/sanity-seo). Run once after wiring sanity-seo's schemas into your Studio.

Install (Phase 4.3 onward)

pnpm add @plesso/content-factory @plesso/sanity-seo

Required peers:

• @sanity/client >=6.22
• @plesso/sanity-seo (any version)

Usage sketch

A minimal Vercel edge function that re-exports the scheduled-run handler:

// api/factory/scheduled-run.js (in your consumer app)
export {default} from '@plesso/content-factory/api/scheduled-run'

Programmatic use of the drafter:

import {draftArticle, commitDraft} from '@plesso/content-factory/lib/draft-article'
import {createClient} from '@sanity/client'

const client = createClient({projectId, dataset, token, apiVersion: '2024-01-01'})
const cluster = await client.getDocument('topicalCluster.anniversary-gifts')

const draft = await draftArticle({
  cluster,
  topic: cluster.queue[0],
  anthropicKey: process.env.ANTHROPIC_API_KEY,
})

await commitDraft(client, draft)

What this package does NOT do

• No Studio UI. The drafter is a backend pipeline; if you want editor-triggered actions (Generate Meta from Body, Compute Readiness, etc.) those live in @plesso/sanity-seo/actions.
• No knowledge intake. The drafter assumes you have a topical cluster + prompt template already populated. The chat agent that produces those is @plesso/knowledge-intake (Phase 5+).
• No methodology routing. Drafter assumes the prompt template encodes whatever style/structure you want. Methodology selection happens upstream in @plesso/methodology (Phase 5+).

License

UNLICENSED for now. Will likely move to MIT alongside the rest of the @plesso/* set once we're ready to open-source the engine layer (separate from the Plesso platform).