10 creative projects for Thota's VPS — AI, automation, and personal clouds, all running lean on 2 vCPU / 4 GB.
Lean VPS forcing lean thinking. No bloat, no enterprise tools, no 2GB Docker containers.
llama3.2 2B via Ollama — the only full LLM that fits alongside everything else. Embedding models for RAG. Whisper.cpp for transcription.
Zero-cost, zero-infrastructure alerting. Python + python-telegram-bot. One bot handles all project alerts.
No daemon RAM overhead. One .db per project. Personal-use concurrent reads are fine. Dead simple backups.
systemd timers trigger heavy jobs. APScheduler only for in-process workflows. Most projects sleep until called.
Flask + vanilla JS + HTMX. No React, no Webpack, no npm overhead. Each Flask app is under 100 lines of core logic.
One VPS, one domain, many sub-paths. babu.thotas.com/papertrail, babu.thotas.com/watchdog, etc. SSL via Let's Encrypt.
Highest cool factor, highest feasibility. These four validate the entire stack before you touch the ambitious ones.
"A self-hosted bookmark & read-later service with AI-powered summarization."
Save links from anywhere. Auto-fetch title, description, and OG image. Get AI summaries so you never waste time on a bad article.
| Metric | Value |
|---|---|
| RAM | ~600 MB |
| CPU | 1 core |
| Storage | 8 GB |
| Cool Factor | 7/10 |
| Feasibility | 5/5 |
| Setup Time | ~2 hours |
"Your own RSS aggregator with daily AI digest emailed to you."
Aggregate all your RSS feeds. Every morning at 7am, a local LLM reads the night's new articles and writes you a crisp digest email.
| Metric | Value |
|---|---|
| RAM | ~400 MB |
| CPU | 1 core |
| Storage | 3 GB |
| Cool Factor | 6/10 |
| Feasibility | 5/5 |
| Setup Time | ~2 hours |
"Voice note transcription hub — drop audio, get text."
Meeting notes, voice memos, voice diary — all searchable. Drop any audio format, get a transcript back. Your voice, your words, instantly findable.
| Metric | Value |
|---|---|
| RAM | ~700 MB |
| CPU | 1 core |
| Storage | 5 GB |
| Cool Factor | 8/10 |
| Feasibility | 4/5 |
| Setup Time | ~2 hours |
"Travel tracker that plots your countries and cities on a map automatically."
Log where you've been. See every city, country, and date rendered on a beautiful world map. No account, no cloud — just your travel history, beautifully visualized.
| Metric | Value |
|---|---|
| RAM | ~300 MB |
| CPU | 1 core |
| Storage | 2 GB |
| Cool Factor | 7/10 |
| Feasibility | 5/5 |
| Setup Time | ~1 hour |
"A personal agent that monitors topics and drafts responses for your review."
Track keywords across RSS, Twitter, Reddit. When something relevant appears, a local LLM writes a draft reply — you review, hit post. Your digital twin that never sleeps.
| Metric | Value |
|---|---|
| RAM | ~800 MB |
| CPU | 2 cores |
| Storage | 5 GB |
| Cool Factor | 9/10 |
| Feasibility | 3/5 |
| Setup Time | ~4+ hours |
10 always-on Flask apps + Ollama doesn't fit in 4GB. Here's the strategy.
WatchDog (200MB) + PaperTrail (700MB) + LinkVault (600MB) + HabitForge (350MB) = 1.85GB
PulseRSS + Nomad + TinySync + PastePit + WhisperBox = loaded only when needed
One Ollama instance. llama3.2 2B = ~1.5GB when loaded. Load on demand, unload after 10 min idle.
| Component | RAM |
|---|---|
| Debian OS | ~500 MB |
| nginx | ~30 MB |
| Ollama (idle) | ~100 MB |
| Always-on apps | ~1.85 GB |
| Total | ~2.5 GB |
| Headroom | ~1.5 GB free |
✅ Fits comfortably. Load Ollama only when a project needs AI.
Per-project virtualenv. One Flask app per project, separate port. Under 100 lines of core logic per project.
One .db per project in /var/data/. No daemon. Fine for personal-use concurrent reads.
The only full LLM that fits alongside everything else on 4GB. Load on demand, unload after idle.
One bot token for all projects. Extremely lightweight. Zero infrastructure cost.
Cron for simple intervals. systemd timers for complex scheduling. No APScheduler daemon.
| Tool | Why |
|---|---|
| n8n | ~1GB+ RAM alone |
| PostgreSQL | Daemon overhead |
| React/Webpack | Build tool RAM bloat |
| Stable Diffusion | CPU-only too slow |
| Django | Heavy for single-user |
10 projects, one VPS, all self-hosted. No subscriptions, no cloud lock-in, no bloat.
Recommended
Always-On
Cron-Triggered
On Demand
Shared Ollama
Service
First project: WatchDog — validates the whole stack in ~1 hour.
then PaperTrail → HabitForge → LinkVault → PulseRSS → Nomad → TinySync → WhisperBox → AutoGrok