55 live n8n workflows. 13 custom AI agents. A Qdrant vector context stack shared across every agent. This is the infrastructure that runs Fatboy Studios and every client engagement — daily, unattended.
Research → brief → draft (Claude API) → design → image generation → Sanity publish. GEO-optimized with per-client backlink index. Min 5 images per post.
Design agent clones Figma template → swaps text + images → exports to R2. 6 QC auto-fix passes per frame before export. Gap logging mandatory.
Portal metrics snapshots at 22:30 SAST daily. Google Ads API, GA4, GSC pulling into Supabase. Sparkline history for every KPI. 13 separate sync workflows.
n8n FanHemNwkwxVtcGV. 6 GraphQL queries → Claude scoring (P1/P2/P3) → ClickUp surface with AI-drafted proposal as comment. Fully automated intake.
Content calendar generation, Doublespeed post scheduling, image variant production. DC + Ryla retainer social automation.
n8n health checks, circuit breakers, error escalation to ClickUp. Uptime Kuma monitoring 15+ services. Grafana dashboards for local infra.
Every agent has a defined scope, a set of MCP tools, and access to the Qdrant context store. Context isolation is hard — each agent only reads what it needs.
CTO — Fatboy
Fatboy infra onlyInfrastructure, deploys, Vercel, Sentry, n8n health, Supabase migrations
CTO — ApeFX
ApeFX onlyApeFX platform deploys, incidents, costs, performance
PM
Full orgClickUp sprint sync, workflow status, daily ops, cross-team routing
CMO
Per clientMarketing strategy briefs, content direction, campaign planning
PPC Specialist
Per clientGoogle Ads API, GA4, bid strategy, keyword expansion, offline conversions
SEO Brief
Per clientBlog brief generation, GEO keyword targeting, AIEO content structure
Social Brief
Per clientSocial post copy, platform-specific formatting, hook library
Design Agent
Design pipelineFigma template-swap, image generation routing, R2 delivery
Competitor Scan
Per clientCompetitive intel gathering, positioning analysis
Client Report
Per clientAutomated monthly reports from portal data
Ads Brief
Per clientGoogle Ads copy generation, task-specific
Campaign Research
Per serviceCampaign concept development, creative direction, 3-concept output
Proof Agent
Content QCFact-check, citation extraction, claim verification
Every agent reads and writes to Qdrant. Per-client collections. Per-domain partitions. Context API on port 3457 — any agent can call GET /context/:clientId to pull relevant state before executing a task.
The difference between an agent stack that 'works' and one that operates in production is persistent memory. Without it, every session starts cold, every task restates the same context, and the system doesn't compound.
Qdrant vector store
Self-hosted Docker. Per-client collections.
Context API (port 3457)
GET /context/:clientId — used by all agents.
Ollama + Qwen local
Local LLM tier for high-volume routine tasks.
Claude API (Sonnet/Opus/Haiku)
Routed by task complexity and cost target.
Agentic systems and automation infrastructure is one of my three engagement types. Tell me what you need to automate.