Skip to main content
Auto-memory
Retell + Twilio
MCP ecosystem
Founder Pres.
Obsidian + QMD
Review pass
Claude / Codex
Convex
75K+ sold
SwiftUI CloudKit
Next.js + Vercel
1,350+ locations
Shopify
Session notes
Founder
MCP forks
<60s response
Operator chapters
QMD retrieval
121+ skills
Six figures raised
Founder office
back to grid
Projects2026

Mortals Automated Marketing

An in-house autonomous marketing system for Mortals. It drafts ads, emails, landing-page copy, and organic posts, then checks each item against budget rules, FDA-safe language, brand voice, and operator approval before anything can ship.

In-house
replaced the paid-media retainer
Claim risk
FDA-safe language before approval
Spend drift
budget rules before platform writes
One queue
human approval for customer-facing work
§01The workLive system

MAM brought Mortals' marketing workflow in-house: drafting, claim checks, budget rules, approvals, and platform writes in one private system.

The work

One private system for Mortals marketing.

Agents draft the work. Convex enforces the rules. Connor approves customer-facing changes from one queue.

Draft -> check -> approve -> writeWorkflow loopOne controlled path from draft to platform write
01Draft
02Check
03Approve
04Write
The workAgents draft the work. Convex enforces the rules. Connor approves customer-facing changes from one queue.
§02The pipelineEnd-to-end flow
01

Draft

Agents draft ads, emails, landing pages, and organic posts from the current brand voice.

AI generation layer
02

Check

The system checks every draft for FDA-safe wording, brand voice, and banned claims.

FDA + brand-voice pass
03

Approve

Connor reviews customer-facing changes in a single queue with diffs and final approval.

Operator queue
04

Write

Approved changes go through typed MCP tools and first-party platform SDKs.

Typed tools + SDKs
§03What I builtPieces that compound

Five pieces make it work: a rules engine, typed tools, review gates, an FDA checker, and direct platform SDKs.

A deterministic rules engine.

Convex enforces platform budget caps, zero-sum reallocations, and a kill switch.

Budget caps · kill switch · zero-sum

Nineteen typed MCP tools.

Agents can only write through typed mam_* tools. Each tool validates input, applies business rules, and performs one controlled side effect.

The only legal write path

An FDA copy checker as its own module.

The checker blocks banned claim verbs and rewrites symptom language before copy reaches the queue.

Banned verbs · required verbs

Direct SDKs into every platform.

Meta, Google, TikTok, Klaviyo, Shopify, and organic channels are wired through first-party SDKs in the same repo as the rules engine.

First-party SDKs
§04The engineRules before writes
Engine guardrailsConvex · unit-tested
  1. Each paid platform has a hard budget cap.
  2. A single breach pauses every ad on every network.
  3. Scaling a winner means cutting a loser on the same platform; any update that does not net to zero is refused.
  4. Customer-facing copy clears the checker, the audit pass, and Connor.
§05ArtifactsReceipts and public proof

MAM is private internal infrastructure, so the runtime isn't public. The architecture is: rules, tools, checks, and approvals.

Deterministic rules + budget engine

Convex modules enforce platform budget caps, zero-sum reallocations, and the kill switch before any budget update can write.

private · Convex + TypeScript
19-tool MCP server

Typed mam_* tools expose ads, budgets, Klaviyo flows, creative review, experiments, and strategy library reads to agents.

private · MCP
Three-tier review + FDA checker

Every customer-facing draft passes generation, audit, FDA language checks, and operator approval before it ships.

private
13-route decision dashboard

Next.js routes give the operator one place to review work, budgets, performance, experiments, tasks, and integration health.

private · Next.js 15 + Convex