Ad-Forge

Ad-Forge is a project-based marketing pipeline at /ad-forge that turns a brief into ads: research → creative → copy → review → publish → analyze.

Ad-Forge - M.A.V.E.N

Purpose

Everything is a project saved in the vault (<vault>/AD-Forge/<slug>/) - the brief, each stage’s output and the research - so a run keeps going even if you switch apps mid-run. Reopen the project and the finished (or still-running) stages are there.

Stages

The pipeline runs six stages: research → creative → copy → review → publish → analyze.

Background jobs - stages run as background jobs; each is an agent step you can re-run and hand-edit.
Per-stage model - pick the model per stage, or leave Default model.
Phase A vs Phase B - Phase A (research → review) spends no money. Publish and analyze are Phase B - locked until you connect an ad account and confirm the spend.

Live web research

With a Tavily or Firecrawl key (.env.local, or the in-app Secrets panel), every model gets live web plus a Search-Picker. Without a key the pipeline still runs, just without live lookups.

Ad accounts

Ad accounts - Meta, Google, YouTube, LinkedIn, TikTok, X, Telegram, Reddit - connect on the right. Their keys are saved to .env.local and masked after saving. Nothing publishes without your confirmation.

Setup and dependencies

TAVILY_API_KEY or FIRECRAWL_API_KEY (optional) - live web research and the Search-Picker. Either provider works; Ad-Forge runs without one. Set via .env.local or the in-app Secrets & keys panel.
Ad-account keys - added per network in the app; saved to .env.local and masked.

Where output goes

Each project lives in the vault at <vault>/AD-Forge/<slug>/ - the brief, every stage’s output and the research.

Tips

Phase A is free - run research through review to build the creative and copy without spending anything.
Publishing and analysis are gated behind a connected ad account and an explicit spend confirmation, so nothing goes live by accident.