Online Optimisers · The Machine
← Back to the orb
// What Donal built · for Mads

One operator. A whole agency's output.

You coach me on the business, so here is the business: the machine I spent 2026 building. Every report in the orb you are holding - the AI probes, the SEO audit, the playbooks - came out of this. One page, all receipts, no fluff.

The year in numbers // all verified, all versioned

1,271
git commits since January
every decision versioned, nothing lives in chat
247
reusable skills
delivery does not slow as clients grow
65
specialist agents
fan out 6-11 at once: 8-person output, 1 head
24
operating rules
quality baked in, written once, run 1000x
~14k
RAG knowledge chunks
a year of distilled frameworks, queryable
158
audits live on one project
extreme capital efficiency
117
tool directories
scrapers, factories, gates, dashboards
9
paying retainer clients
the machine runs a real book, today

Always-on automations // runs while I sleep

AutomationWhat it doesCadence
AEO citation loggerRe-asks the AI engines the money questions and logs who gets cited, per clientWeekly, scheduled
Cost dashboardTracks every API dollar across the stack; pre-flight gate blocks runaway spend before it happensDaily
Web QA watchWatches client sites for breakage (forms, renders, redirects) so a silent failure never runs for weeksScheduled
Git auto-backupCommits and pushes all work automatically; nothing is ever lostEvery ~15 min
Voice drainVoice notes auto-transcribed and routed into the knowledge systemContinuous
RAG retrieval crons6 retrieval jobs that surface relevant knowledge before work starts (zero API cost by design)Drafted, loading

n8n workflow automation // the no-touch pipelines

Lead Strike

A cold lead hits the list and the machine fires: webhook in, audit generated, personalized assets out. No human in the loop until a reply.

Instantly reply relay

Email replies from outreach campaigns route instantly to the right place; no inbox-watching.

Lead list builder

Builds qualified lead lists from raw sources, enriched and formatted for campaigns.

Monthly data pull

Client performance data collected on schedule, feeding the monthly reports without manual exports.

webhook: new lead crawl + audit personalized video (VSL render worker) landing page email draft (human sends)

The VSL pipeline: a prospect's name goes in, a personalized video + GIF thumbnail + landing page + email draft come out in ~5-8 minutes. Plus an image factory and video factory for branded creative at scale. Every outbound email is human-approved by rule - the machine drafts, Donal sends.

The connected stack // APIs wired in

Data + SEO intelligence
DataForSEO Google Search Console Google Analytics 4 Google Business Profile Lighthouse / PageSpeed Firecrawl

AI engines
Anthropic / Claude OpenAI Perplexity

Delivery + infrastructure
Cloudflare Pages + Workers Resend (transactional email) Supabase (vector database) GitHub Google Sheets / Drive

Sales + ops
Instantly (outreach) Apollo Hunter Airtable (ops hub) GHL (webhooks, per-client) Fathom (call transcripts)

~20 services wired into one orchestrator. The point is not the logos: it is that an audit, a report, or a campaign can pull real data from all of them in one run - like the probe that produced your AI Visibility report for $0.32.

Report + audit engines // what produced your orb

EngineWhat it producesProof
AI visibility probeAsks ChatGPT + Perplexity the buyer questions, logs who gets cited, names the gapYour 0/11 finding, run live, $0.32
Full auditCrawl + rankings + backlinks + AI citations + schema, one styled deck158 of them live on one project
AI audit deepEvery signal that decides whether AI recommends a businessThe quarterly engine for retainer clients
Deep audit5-tool synthesis: ranking-drop forensics, competitor gap, CWVDiagnosed a client's 3-cause ranking collapse
Monthly reports6 data sources per client, batched, plain-English narrative10 clients in ~9 minutes, ~$2.70 total
Content gap + strategyAI content-gap analysis, refresh specs, 90-day strategy mapsThe suite in this orb, and the one a SaaS founder got last month

The knowledge brain // RAG: a year of learning, queryable

3
knowledge vaults
courses + intel / agency ops / personal
~14k
indexed chunks
hybrid vector + keyword retrieval
10+8
skills + SOPs wired to it
retrieval is a reflex, not an afterthought

Every course, podcast, framework, and client lesson from the last year is distilled, embedded, and searchable by the machine itself. When a skill runs, it first asks "what do we already know about this?" - and every retrieval is graded afterward (helpful or not), so the brain gets sharper with use. Built as a custom MCP server on a vector database.

Why it matters to you: your DISC and delegation IP could work the same way - one structured brain that your content, your courses, and one day your own AI assistant all draw from. That is the pattern behind the GEO playbook in this orb.

How it runs // the operator OS

Donal + Claude Code (the orchestrator) 247 skills+ 65 agents in parallel+ 24 rules+ the RAG brain audits, reports, sites, campaigns
What this is not (the honest bit)

The 65 agents are orchestrator-run, not autonomous employees - the leverage is structured parallel dispatch, not magic. The numbers above are real and versioned, but the machine still needs its operator: judgment does not delegate. That, as it happens, is exactly what you coach me on.

And this page exists because you coach me on the business. It is a show-and-tell between friends, not a brochure.

Online Optimisers · built for Mads Singers · 2026-06-12 ← the orb · the short version · the GEO playbook