Building in Public

I Taught My Laptop to Work the Night Shift

How I built 6 compounding AI automation tools on a regular Windows laptop in a single afternoon. They run while I sleep. They cost nothing. And they get smarter every night.

📝 Will Tygart 📅 March 21, 2026 ⏱ 8 min read 🛠 Claude + Ollama + PowerShell
Scroll to explore

The Problem: Too Many Sites, Not Enough Hours

I manage 23 WordPress sites across multiple businesses. Content needs refreshing. SEO needs monitoring. Meeting notes need processing. Briefs need writing. And I have exactly one set of hands.

23
WordPress Sites
6
Businesses
468
Files to Index
$0
Monthly API Cost
The question wasn't "can I afford enterprise automation software?" It was "what if the laptop sitting on my desk could do this work itself, for free, while I'm asleep?"

Most people think you need Zapier, Make, n8n, or some cloud workflow engine to automate business operations. You don't. You need a laptop, a local AI model, and a willingness to think differently about what your hardware can do.

The 6 Tools I Built in One Session

Each tool is a standalone PowerShell script that calls Ollama's local REST API. No Python dependencies. No virtual environments. No package conflicts. Just scripts that work.

📡
Site Monitor
Every hour
Pings all 23 sites, measures response times, logs to CSV. Windows notification if anything goes down. Caught a 3.8-second response time on one site during the first run.
📜
Nightly Brief Generator
2:00 AM daily
Reads from a topic queue, generates JSON content briefs via Ollama. Produces 5 publish-ready briefs per night. 150 briefs per month, zero human effort.
🎙
Meeting Processor
6:00 AM daily
Finds Gemini transcript files, extracts key decisions, action items, follow-ups, and notable quotes. Never lose a detail from a client call again.
🔍
Auto-Indexer
3:00 AM daily
Scans all text files, generates 768-dimension vector embeddings via nomic-embed-text. Enables semantic search across every document, brief, and draft. Gets smarter every night.
📧
Email Digest
6:30 AM daily
Scans Outlook and local email files, categorizes by priority level, generates a morning briefing. Wake up knowing exactly what matters.
📈
SEO Drift Detector
7:00 AM daily
Checks all 23 sites for title tag, meta description, H1, and status code changes. Alerts on any drift. Catches problems before Google does.

What My Laptop Does While I Sleep

Every night, this sequence fires automatically via Windows Scheduled Tasks. By the time I pour my coffee, the work is done.

2:00 AM
Brief Generator wakes up
Pulls 5 topics from the queue. Ollama generates structured JSON content briefs with target keywords, audience segments, and content angles. Marks them complete.
3:00 AM
Auto-Indexer starts scanning
Finds every new file created since yesterday. Generates 768-dimension vector embeddings. Updates the search index. The briefs from 2 AM? Already searchable.
6:00 AM
Meeting Processor runs
Scans for any new Gemini transcripts. Extracts decisions, action items, and follow-ups into clean markdown summaries.
6:30 AM
Email Digest compiles
Categorizes yesterday's emails into Urgent, Client-Critical, Opportunities, and FYI. Generates a prioritized morning briefing.
7:00 AM
SEO Drift Detector sweeps
Checks all 23 sites for any changes to titles, meta descriptions, or H1 tags since yesterday. If something changed, a Windows notification is waiting.
Every hour
Site Monitor pings everything
All 23 sites checked. Response times logged. If anything goes down, a balloon notification appears immediately. 24/7 coverage.

First Run Results

Every tool was tested live during the build session. Here's what happened on day one.

20/23
Sites Responding
48
Files Indexed
768
Vector Dimensions
~33s
Per Generation
# Site Monitor - First Run Output
Sites checked: 18/18
Sites UP: 17
Sites DOWN: 1 (403 - firewall, not actual downtime)
Slowest: mintcomedy.com - 3.8s
Fastest: borro.com - 0.4s

# SEO Drift Detector - First Run
Sites scanned: 23
Successfully checked: 20
Baseline created: Title tags, meta descriptions, H1s, canonicals
Drift detected: None (first run = baseline)

# Vector Indexer - Embedding Test
Model: nomic-embed-text
Dimensions: 768 per file
Files queued: 468
Status: Running in background...
The site monitor found a performance issue on its very first run. Mint Comedy was responding in 3.8 seconds — nearly 10x slower than average. That's the kind of thing that tanks Core Web Vitals and you'd never notice manually.
— Caught in the first 30 seconds of monitoring

The Entire Stack

No cloud subscriptions. No API keys. No monthly bills. Everything runs on hardware you already own.

🤖
Claude (Cowork Mode)
Built all 6 tools in a single session via Claude's computer use. Controlled the actual Windows machine through MCP.
🦙
Ollama
Local LLM server running llama3.2:3b for generation and nomic-embed-text for vector embeddings. REST API on localhost.
PowerShell
Every script is pure PowerShell. No Python installs, no package managers, no dependency hell. Works on any Windows machine.
🕒
Windows Task Scheduler
Native OS scheduling. No cron, no daemon, no third-party tools. Just set the time and forget it.

Cost Comparison

Zapier Pro (multi-step automations)$69/mo Make.com (10k operations)$16/mo n8n Cloud Starter$24/mo UptimeRobot Pro (50 monitors)$7/mo ContentKing (SEO monitoring)$49/mo OpenAI API (equivalent usage)~$30/mo Comparable SaaS total$195+/mo This laptop stack$0/mo

The Compounding Effect

These tools don't just run in parallel. They feed each other.

The Brief Generator creates content briefs at 2 AM
The Auto-Indexer picks them up at 3 AM and makes them searchable
The Meeting Processor extracts action items that feed back into the brief queue
The SEO Drift Detector catches changes that trigger new monitoring entries
The Vector Index grows every night, making semantic search across all outputs more powerful
Every tool I add makes every other tool more useful. The vector index gets richer. The brief queue gets smarter. The monitoring gets more comprehensive. This is what compounding automation actually looks like — not a workflow that runs once, but a system that improves itself.
Your Laptop Is Underemployed

The hardware sitting on your desk right now can run local AI models, automate your operations, and work while you sleep. It just needs someone to show it how.

Explore Tygart Media →