Skip to content

Work · Growth infrastructure / Automation

Growth OS

Signals → guardrails → decisions → actions → measurement.

What it is

Growth OS is an internal marketing operating system. It turns real product signals into campaign strategy and channel-specific content, runs each draft through a deterministic claims gate, requires human approval, packages publication with attribution, and measures and learns from the result.

Why it needed to exist

A small studio cannot staff a marketing team, and generic AI content invents claims. The system had to reduce the founder's workload while preserving judgment and refusing to publish anything unsupported.

What Idealink did

Idealink designed and built the system as a modular monolith with a governed event pipeline, and connected its first product, Declora, through a versioned adapter.

Status
In development
Relationship
Idealink venture
Industry
Marketing operations
Period
2026 — internal
Platform
Web · API · Node SDK
Idealink role
Architecture, engineering, product
SIGNALS → GUARDRAILS → DECISIONS → ACTIONS → MEASUREMENTSIGNALSsanitised eventsSTRATEGYapproved briefsCONTENTfrom product memoryCLAIMS GUARDdeterministicHUMAN APPROVALapprove · reject · changesPUBLISHpackage · attributionMEASUREevents · learningGROWTH OS PREPARES · HUMANS DECIDE · NOTHING AUTO-PUBLISHES
System diagram

Thesis

Marketing automation fails when the machine is allowed to say things the product cannot back. Growth OS is the decision-loop story: product signals in, campaigns and content out, every claim checked deterministically, every publication approved by a human, everything measured.

01

The loop

Eight stages, implemented end to end for the first connected product.

  1. Product signals

    Sanitised, aggregated observations arrive through an event gateway with a schema registry — never raw customer data.

  2. Campaign strategy

    Signals become briefs; a brief needs approval before it can spawn content.

  3. Content

    Channel-specific drafts are generated from an approved product memory, the single source of product truth.

  4. Claims Guard

    Deterministic checks: evidence-required claims warn when unsupported; prohibited claims block until corrected or overridden with a reason, a user and a timestamp.

  5. Human approval

    Approve, reject or request changes. The system prepares; humans decide. Nothing auto-publishes.

  6. Measure and learn

    Attribution links and a canonical event model feed measurement back into the next brief.

02

Architecture

Shape
pnpm workspace modular monolith: one deployable Next.js app plus packages for domain, intelligence, claims guard, attribution, protocol, event catalog, normalisation and integrations
Protocol
CloudEvents envelopes, privacy classification, API keys, idempotent delivery
Integrations
LLM providers behind a provider layer; a Node SDK and a versioned adapter for connected products
First product
Declora, an app-submission product, connected through a frozen v1 contract
Deployment
Vercel and Neon Postgres; unit, integration and end-to-end tests

Growth OS prepares; humans decide. Model output never becomes a public claim without evidence and a signature.

Outcome

Where it stands

The full loop runs internally for Declora — signals, briefs, guarded content, approval, publication packages and measurement — and is the reference for how Idealink applies language models inside governed workflows.

Internal system. No customers, volumes or channel results are published.

Stack and capabilities

Only what mattered

  • Next.js
  • TypeScript
  • Prisma
  • CloudEvents
  • LLM providers
  • Vercel · Neon

Next · Consumer product / Privacy-first

Ownword

A private ledger for tiny daily promises.