I Cut My Tool Subscriptions in Half. A VPS and Two Open-Source Projects Replaced Them.
Between Vercel, ChatGPT Plus, GitHub Copilot, and a handful of SaaS tools, I was spending over $60/month on subscriptions. I moved to a VPS, Dokploy, and Hermes Agent — and I'm never going back.
I added up my subscriptions last month. Between Vercel Pro, ChatGPT Plus, GitHub Copilot, a monitoring tool I forgot I was paying for, and a few other things that auto-renewed without me noticing — I was at $74/month. For tools. Just tools.
That’s $888 a year. On subscriptions I didn’t think twice about because they were all $10-20 each.
I didn’t cancel everything at once. But over the last six months, I’ve replaced most of it with three things: a Hetzner VPS, an open-source deployment platform called Dokploy, and an AI agent called Hermes. My monthly bill is now around $25. And honestly? The replacements are better.
The VPS nobody told me I needed
I’ve been renting a Hetzner VPS for a while — 4 vCPUs, 8GB RAM, 80GB SSD, €4/month. I used it for random projects, nothing serious. It sat there mostly idle while I paid Vercel $20/month to host two static sites and a small API.
That math is absurd in retrospect.
The problem was never the VPS. The problem was that deploying to a VPS felt like work. Docker Compose files. Nginx configs. SSL certificates. Environment variables scattered across .env files. I knew how to do all of it, but I didn’t want to. Vercel charged $20/month for the same reason Uber charges for rides — not because driving is hard, but because not thinking about parking is worth money.
Then I found Dokploy.
Dokploy: the Vercel experience on your own hardware
Dokploy is an open-source PaaS — a deployment platform you host yourself. You point it at a Git repo, it builds and deploys your app. It handles SSL, environments, rollbacks, monitoring. It looks like Vercel. It works like Vercel. But it runs on your VPS.
The first time I deployed a Next.js app to Dokploy, I was waiting for the catch. There wasn’t one. It pulled the repo, built the Docker image, provisioned SSL through Let’s Encrypt, and gave me a URL. Same flow as Vercel. Zero Nginx configs. Zero certbot commands.
It also handles databases. Postgres, MySQL, MongoDB, Redis — one-click from the dashboard. No separate cloud DB subscription. No connection string juggling. The database runs on the same VPS as the app.
Here’s what my deployment costs look like now:
| What | Before | After |
|---|---|---|
| Hosting (2 sites + API) | Vercel Pro: $20/mo | Included in VPS |
| Database | Supabase Pro: $25/mo | Included in VPS |
| Monitoring | $9/mo | Dokploy built-in |
| Total hosting | $54/mo | €4/mo |
The VPS handles all of it. CPU sits at 15%. RAM at 30%. There’s room to spare.
Dokploy isn’t the only option — Coolify and CapRover do similar things. But Dokploy’s UI is the closest to the Vercel experience, and the Docker-based architecture means you can deploy anything, not just JavaScript frameworks.
I’m presenting Dokploy at the Nairobi Linux User Group next month. More on that later.
Hermes: the AI assistant that doesn’t cost $20/month
The other subscription that bothered me was AI. ChatGPT Plus at $20/month for GPT-4 access. GitHub Copilot at $10/month. Between the two, $30/month for two models that sometimes refused to answer perfectly reasonable questions.
I switched to Hermes Agent, an open-source AI agent framework, connected to DeepSeek’s API through OpenRouter. DeepSeek V3 costs about $0.27 per million input tokens. For comparison, GPT-4 is $30 per million. That’s 100x cheaper.
My monthly API bill on DeepSeek: $8-12. For everything. Coding assistance, research, writing help, server automation. It’s not GPT-4, but for most tasks I don’t notice the difference. When I need frontier-level reasoning, I can switch models mid-conversation without changing anything else.
But the cost savings aren’t why I stayed. I stayed because Hermes remembers.
Every conversation builds on the last one. It knows I’m a Python developer targeting remote roles. It knows I prefer short, direct answers. It knows my file structure, my naming conventions, my writing voice. ChatGPT resets every session. Hermes accumulates.
Last week I asked it to help me set up a Dokploy deployment for a Django app. It didn’t need to ask what version of Python I use. It didn’t need to ask where my projects live. It knew because I’d told it, once, months ago. That’s not a chatbot. That’s a collaborator.
The stack I’m running now
Everything lives on one €4/month Hetzner VPS:
Hetzner VPS (Ubuntu, 4 vCPU, 8GB RAM)
├── Dokploy → deploys my apps and databases
├── Hermes Agent → AI assistant, runs 24/7
│ ├── Telegram bot → chat from my phone
│ └── Hermes Desktop → native GUI via Tailscale
├── Postgres + Redis → managed by Dokploy
└── 2 Next.js sites → deployed via DokployTotal monthly cost:
| Item | Cost |
|---|---|
| Hetzner VPS | €4.00 |
| DeepSeek API (via OpenRouter) | ~$10.00 |
| Tailscale (free tier) | $0 |
| Total | ~$14.00 |
Previous monthly subscriptions: $74. Current: $14. Savings: $60/month, $720/year.
And I haven’t lost anything. My sites are faster because the database is local. My AI assistant is more useful because it has persistent memory. My deployments are one git push away. Nothing depends on a venture-backed startup that might change pricing or shut down.
The philosophy behind all of this
There’s a pattern here that goes beyond cost.
Dokploy replaces Vercel. Hermes replaces ChatGPT. Both are open-source. Both run on hardware I control. Both get better the longer I use them because they accumulate my data, my preferences, my context.
This isn’t about being cheap. It’s about owning the things that make you productive.
When Vercel changes its pricing, I’m not affected. When OpenAI adds restrictions to GPT-4, I’m not affected. My stack doesn’t depend on any single company’s goodwill or business model. It depends on a VPS, some open-source projects, and an API key.
The API key part is worth addressing. I’m still using a cloud LLM — DeepSeek through OpenRouter. If DeepSeek disappears tomorrow, I swap the API key for another provider. Hermes supports 20+ providers. The model is a commodity. The agent, the memory, the accumulated knowledge — that’s irreplaceable, and it lives on my server.
What I’m presenting at Nairobi LUG
Next month I’m giving a talk at the Nairobi Linux User Group about Dokploy: what it is, how to set it up, and why self-hosting your deployment platform is worth the initial effort. I’ll also touch on Hermes and the broader theme of building a personal infrastructure stack that doesn’t depend on subscriptions.
If you’re in Nairobi, come through. If you’re not, I’ll post the slides and a recap.
In the meantime: look at your subscriptions. Add them up. Ask yourself which ones you’re paying for because the alternative felt like work. The alternative probably got easier since you last checked.
This is part of my AI Agents for Daily Life series. Previous: I Didn’t Get the AI Agent Hype. Next: contributing to Dokploy with an AI agent as my pair programmer. Follow at nyagah.me or find me on X and LinkedIn.