Real work. Real results.
Case studies
Three levels of automation — from individual workflow to fully autonomous company. All built and operated by one person.
Process Automation
Automating business operations with n8n
n8n is the backbone of every process that used to require manual effort — analytics reports, content pipelines, approval flows, payment notifications, newsletter dispatch. The goal was to make recurring operations run without touching them.
By the numbers
23
active workflows in production
8
sites generating content daily
0
manual steps in daily operations
€0
SaaS tools replaced by self-hosted n8n
What was automated
Daily analytics reports
Umami traffic data for 8 sites consolidated into one Telegram message every morning at 08:30. Replaced manual dashboard checks and 8 separate report jobs.
SEO keyword research pipeline
Runs daily per site: fetches live data (NHL API, seed topics), scores topics via LLM, saves approved keywords to Minio. Content generator reads the file at next run.
Consulting inquiry flow
Form submission → Telegram approval with inline buttons → automated email with Stripe payment link → Telegram confirmation. No calendar exposure, no manual steps.
LLM spend monitoring
Weekly report: total spend, per-model breakdown, daily trend — pulled from LiteLLM API and sent to Telegram Reports topic every Monday.
Newsletter dispatch
Every published article triggers Resend email dispatch to subscriber list + Bluesky post automatically. Zero post-publish manual work.
Error monitoring
Any workflow failure triggers a Telegram alert with workflow name, node, and error. Catches issues without log watching or uptime dashboards.
Stack
n8n (self-hosted) · Groq / Gemini / Anthropic via LiteLLM · Minio (S3) · Postgres · Telegram Bot API · GitHub API · Resend · Umami · IndexNow
"The turning point was realising that n8n isn't a 'Zapier replacement' — it's a runtime for autonomous processes. Once each workflow can receive a webhook, store state in Minio, and send a Telegram message, you have a programmable ops team."
Total Content Automation
Publishing at scale with OpenClaw
OpenClaw is the publish layer — a self-hosted webhook server and script runtime that takes a content draft and handles everything from there: write the file, commit to GitHub, trigger Netlify/Vercel deploy, notify Telegram, and dispatch the newsletter. n8n generates the content. OpenClaw ships it.
By the numbers
8
sites publishing via a single OpenClaw instance
Daily
publishing cadence — every site, every day
<30s
draft → live on site (including Netlify build)
0
humans involved in publish flow
How the pipeline works
Keyword research — n8n workflow fetches live data (NHL standings, seed topics, performance insights) and scores topics via LLM. Approved topics saved as JSON to Minio.
Article generation — n8n generator workflow reads keywords, calls LiteLLM (Gemini Flash, Groq) with a site-specific system prompt, runs optional quality gate, and sends draft + metadata to OpenClaw via webhook.
OpenClaw publishes — writes draft to disk, commits to the site's GitHub repo, Netlify/Vercel auto-deploys on push, IndexNow pings search engines, Telegram notifies, n8n triggers newsletter dispatch.
Optional approval gate — for EIK (emilingemarkarlsson.com), content goes through Telegram inline buttons before publishing. THA and THB auto-publish. No other human involvement anywhere.
Sites running on this pipeline
Stack
OpenClaw (self-hosted Python webhook server) · n8n · LiteLLM · Gemini 2.5 Flash · Groq · GitHub API · Netlify / Vercel · IndexNow · Astro / React / Vite · Resend · Listmonk
"The key insight: separating generation (n8n) from publishing (OpenClaw) makes both sides replaceable. I can swap the LLM, change the content format, or add a new site without touching the publish infrastructure."
Full Company Automation
Paperclip — an AI venture studio that runs itself
Paperclip is the operating layer above the workflows — a persistent AI agent platform where each agent has a role, a workspace, memory, and a heartbeat. CEO, CFO, CMO, and domain-specific analysts run on schedule, read their context from AGENTS.md files, and post decisions and recommendations to Telegram. The company runs even when I don't.
What's running
C-suite
CEO + CFO agents with weekly heartbeats
Domain
analytics, SEO, and content agents per site
~€15
per month total LLM spend across all agents
Self-hosted
Hetzner VPS — no SaaS agent platform costs
How agents are structured
Weekly heartbeat — strategic overview
Reads AGENTS.md containing: org chart, all 8 sites, LLM budget, consulting status, revenue streams. Runs weekly, generates a strategic pulse report, flags risks and opportunities — posted to Telegram Reports topic.
Weekly heartbeat — financial tracking
Tracks: Stripe revenue per service (2,500 kr discovery, retainer, project), LiteLLM API spend (budget: $15/month), gross margin (~98% target). Weekly financial snapshot with spend vs. runway context — posted to Telegram Reports.
Daily — traffic analysis and content feedback
One agent per site cluster — reads Umami page view data, identifies top-performing articles, detects traffic drops or spikes, and feeds insights back into the keyword research pipeline as performance context for next-day generation.
What "runs itself" actually means
Fully autonomous
- → Daily content for 7 of 8 sites (no human input)
- → Analytics reports and weekly performance digest
- → LLM spend monitoring and budget alerts
- → Error detection and Telegram alerting
- → Newsletter dispatch after every publish
- → CEO + CFO weekly reports
Human approval kept
- → EIK content — Telegram button before publish
- → New consulting inquiry — approve before email
- → Pushing this repo (tur-coolify-setup) to GitHub
- → Changing secrets or Coolify env vars
- → New paid services or infrastructure changes
Stack
Paperclip (self-hosted AI agent platform) · PGlite (embedded Postgres) · LiteLLM · Coolify on Hetzner · n8n · OpenClaw · Minio · Telegram Bot API · Stripe
"Paperclip changes the mental model from 'I run workflows' to 'I run a company'. The agents have roles, context, and recurring responsibilities — not just triggers and actions. The difference is that agents can reason about what they're doing, not just execute steps."
Want this for your business?
I can build the same infrastructure for you — from a single automation to a full agent stack. Start with a 30-minute conversation.