Marketing OS
3 agents built, 7 proposed, reading 11 knowledge kinds, across 32 workflows in 5 verticals. Every node opens its spec.
Blog pipeline
Source article to published post. Two compliance checks, one human gate.
- Sourcescrawler · Tavily · GNews · Apify
- Discovery Agentsummary · category · geo · scores
- Feedbrowsable, scored
- Compliance Agentsubject: "source" — safe to write from?
- Creative Agentdraft · repair · humanize
- Draftstops at "ready"
- Compliance Agentsubject: "draft" — safe to publish?
- CMS Gatehuman acknowledges high severity
- CMShuman clicks send
Scripts and social
A different tail — the calendar, not the CMS. Nothing publishes from it yet.
- Approved scriptfrom Creative Agent
- Scheduler Agentslot · spacing · cadence
- Calendarcontent_calendar_entries
- Publishunbuilt — still manual
Intelligence layer and agents
The layer is the authority. Agents reason over it and cite what they used.
One shared store of what the business knows. Agents read from it and cite what they used — the store is the authority, not the model.
every agent reads from it
What a scraped article says, is worth, and belongs to
Whether text is safe to write from, or safe to publish
How to say it — blogs, scripts, hooks
Whether a vernacular rendering keeps meaning, tone and compliance safety
What the team's own choices imply about what to surface next
Whether a mention is a real reputational issue, and what to do
Which template and theme fit, and whether output clears guard rails
Which market × topic combinations deserve a page
Which piece takes which calendar slot — mostly rules, narrow judgment
Which cohort, which trigger, which moment
Workflows
Status is read from each spec, not restated here.