All case studies

AI SaaS · Newsroom OS

Beatwire

Signal to published news across the web in one automated loop.

An AI newsroom OS that turns scattered signals into published, human-approved stories — multi-model drafting, AI editorial review, and one-click multi-channel publishing. It also powers the Content Agent in this dashboard.

In DevelopmentFeature-complete MVP · private beta-ready

Beatwire

AI Publishing

By the numbers

~46k

Lines of code

90

API routes

22

Database tables

4

LLM adapters

The challenge

Independent publishers spend hours monitoring disaggregated channels, drafting breaking summaries, formatting imagery, and re-posting across CMS platforms. Generic AI writers produce ungrounded, single-paragraph text with no editorial controls and no path to publication.

The solution

An end-to-end, multi-tenant Next.js pipeline: Radar ingests and clusters signals, the Story Builder drafts and refines across models, the Editorial Desk runs an AI review and score, and Publish pushes the finished story to multiple destinations — with a human approval gate before anything goes live.

beatwire · pipeline.ts
// Beatwire · signal → published story
radar.cluster(signals)              // group breaking items
  .draft({ models: 4 })             // multi-model draft
  .review()                         // AI editorial score
  .awaitApproval()                  // human gate — required
  .publish(["wordpress", "ghost"])  // multi-destination

Illustrative: a signal becomes a published story through review — with a required human approval gate before anything goes live.

Technical deep-dive

01

Architecture

Next.js 15 App Router, React 19, TypeScript, Supabase PostgreSQL with a 22-table schema, Clerk multi-tenant auth, and Vercel serverless functions backed by 6 background cron workers.

02

AI engine

A provider-agnostic adapter connects Anthropic Claude, OpenAI, Google Gemini, and Venice AI, allowing per-draft model switching without touching the pipeline logic.

03

Ingestion

An SSRF-hardened parser handles RSS auto-discovery, the Reddit API, and YouTube transcript extraction, then clusters related items into coherent story candidates.

04

Publishing

Direct API publishing into WordPress REST, the Ghost Admin API, Framer webhooks, Resend newsletters, and native multi-tenant hosted sites.

Product decisions

  • Human review before publish — the AI drafts and scores, but a person approves.
  • Provider-agnostic model layer so no single vendor lock-in slows the newsroom.
  • Multi-tenant from line one so teams and solo editors share the same secure core.
  • Feature flags gate regulated or unfinished destinations until they are verified.

Built with

Next.js 15React 19TypeScriptSupabase / PostgreSQLClerkVercel CronAnthropic ClaudeOpenAIGoogle Gemini

Honest statusIn active development. Feature-complete MVP that is private beta-ready.

Related guides

Want something like this?

Tell us what you’re building, the code you have, or the problem you need solved.

Start a project
Beatwire — AI Publishing — Kruzeniski.ai · Kruzeniski.ai