The Exact AI Automation Stack I Use as a Solopreneur in 2026 (With Real Monthly Costs)

You are paying $20 here, $29 there, and you have still hit your n8n Cloud execution cap before the month ends. This entire stack, the one that runs this blog and automates my one-person business, costs under $10 a month, and the breakdown is below.

The problem is subscription creep. You signed up for n8n Cloud Starter because self-hosting sounded complicated. You added a ChatGPT Plus seat "for drafting." You kept the Notion AI add-on. Each line is defensible alone, but together they add up to $60 to $100 a month for a business that could run on a tenth of that if you picked tools by the job, not by the landing page.

I rebuilt the stack from scratch in early 2026 with one rule: every layer earns its cost or it goes. Here is what survived.

Each layer of the stack does one job. Overlap means you are paying twice.

What Does a Real Solopreneur AI Automation Stack Cost Per Month?

My all-in cost is roughly $7 to $9 a month. That covers every layer of a working content business: site hosting, database, CDN, version control, email delivery, workflow automation, and AI model access. The two paid lines are a VPS and metered Claude API tokens. Everything else is a deliberate free-tier choice.

The table below maps each layer by job. Read it left to right: what I need done, what tool does it, what tier I run, and what I actually pay.

That is the full bill. No hidden seats, no annual contracts buried in the footnotes.

How I Run Automation Without n8n Cloud Fees

Self-hosting n8n on a Hetzner CX22 VPS is the single best switch in the stack. The CX22 costs about $4.90 a month (as of mid-2026 pricing) for 2 vCPUs and 4 GB of RAM. It runs n8n plus a Postgres database in Docker, and it has never been the bottleneck for any workflow I have thrown at it.

The n8n Cloud Starter is $20 a month and caps you at roughly 2,500 executions. At a moderate publishing cadence, that cap lands in the third week of the month. Self-hosting removes per-execution fees entirely. The only ceiling is the CPU and RAM on your box, and the CX22 handles everything a one-person business needs.

I picked Hetzner over DigitalOcean because the CX22 is consistently cheaper for equivalent specs, and the datacenter coverage (US, EU, Asia) means low latency wherever my workflows call external APIs.

A $4.90 VPS does the same job as a $20 Cloud plan, without execution caps.

Which AI Models I Actually Use (and When Local Beats the API)

For quality-critical output, I use the Claude API. For everything else, I run Ollama locally on a 16 GB MacBook. My metered Claude spend is $1 to $3 a month at my current publishing volume, because I only call the API when it genuinely changes the output quality.

The split in practice:

• Claude API: long-form post drafts, structural editing passes, any output that gets published or sent to a client
• Ollama (local): classifying RSS items, extracting metadata from PDFs, summarising private documents, bulk reformatting tasks

Ollama is MIT-licensed and free. The models I use most (Llama 3 variants, Mistral) run comfortably in 8 GB of RAM, so the 16 GB MacBook is never stressed. Marginal cost: $0. The hardware was already owned.

If you are still deciding between local model runners, I compared the main options in detail: Ollama vs LM Studio vs Jan: Best Local LLM in 2026.

The Claude API is usage-based with no monthly seat. A seat-based plan charges you whether you used it or not. Metered pricing means a slow publishing month costs less, not the same. At my volume, the difference between $20/month (seat) and $1 to $3/month (metered) is $17 to $19 a month, roughly $200 to $228 a year.

How the layers connect: n8n is the orchestration spine; everything else is a node it calls.

Where I Host the Site and Store Data for $0

Four tools handle the entire infrastructure layer for this blog at zero monthly cost: Vercel, Supabase, Cloudflare, and GitHub. Each is genuinely free at solo-operator volume, not a loss-leader trial.

Vercel Hobby hosts the Next.js site. It is free for non-commercial personal projects, with automatic SSL, global CDN, and preview deployments on every push. Supabase Free gives me a shared Postgres database (500 MB), 1 GB of file storage, and up to 2 active projects. Cloudflare Free handles DNS, unmetered CDN, and a basic web application firewall. GitHub Free covers unlimited public and private repos for individual accounts.

Resend handles transactional and newsletter email. The free plan allows 3,000 emails a month and 100 a day. At my current list size, I have never come close to that ceiling.

The domain runs through Namecheap. A standard .com renewal is $11 to $15 a year depending on the promo, which amortises to roughly $1 a month. That is the only infrastructure line that costs anything, and it would cost the same at any registrar.

Two paid lines, seven free. This is what a deliberate stack looks like.

The Paid Lines, and Why They Are the Only Two Worth It

Two lines carry a real monthly cost: the Hetzner VPS at $4.90 and the Claude API at $1 to $3. Every other layer is either free or runs on hardware I already own.

The VPS is not optional if you want automation without execution caps. It also hosts Postgres, so Supabase stays on the free tier instead of needing the $25 Pro plan. One $4.90 box does the work of two paid subscriptions.

The Claude API is not a seat. I do not pay $20 a month for a Claude.ai Pro subscription because I do not need the consumer interface. I call the API directly from n8n, pay per token, and get the same model quality.

For building and editing the site code itself, I use a free AI IDE. I covered the best options in Best Free AI IDE in 2026 if you want the comparison.

How Solopreneurs Overspend on Their AI Automation Stack

The most common mistake is paying for a seat when metered access exists. ChatGPT Plus at $20 a month is a fixed cost whether you use it 30 days or 3. The Claude API charges only for what you call. For an automation-first workflow, metered nearly always wins.

The second mistake is never self-hosting. The n8n Cloud Starter is $20 a month for a capped product. A $4.90 VPS runs the same software without caps. The barrier is one afternoon of setup, not ongoing complexity.

Third: using a frontier model for every task. Running a Llama 3.1 8B model locally via Ollama for classification or extraction is free. Calling Claude for the same task burns tokens unnecessarily. Match the model tier to the actual quality requirement of the job.

Fourth: tool overlap. Stacks with Zapier AND Make.com AND n8n, each handling different automations because the owner added them one at a time without auditing. Pick one orchestration layer and migrate everything to it.

A concrete example of what you can automate for under $5 a month in API costs: How Solopreneurs Use AI to Automate Lead Follow-Up.

Audit for overlap before adding a new tool. Most stacks have at least one redundant pair.

How to Copy This Stack This Week

Start with two decisions that cascade everything else: pick your hosting layer and pick your automation layer. Here is the order that minimises rework.

1. Register your domain on Namecheap (or transfer an existing one). Point nameservers to Cloudflare immediately, even before the site exists. Cloudflare is free and takes 5 minutes to set up.
2. Spin up a Hetzner CX22 VPS. Install Docker and Docker Compose. This is your automation and database server from day one.
3. Deploy n8n via Docker on that VPS with a Postgres backend. Follow the self-host guide for the exact compose file and reverse-proxy config.
4. Create a free Supabase project for your site's database and file storage. Connect it to n8n with the Supabase node.
5. Deploy your Next.js site to Vercel Hobby. Connect your GitHub repo. Vercel handles CI/CD automatically on every push.
6. Add Resend for email. Drop the API key into n8n for any workflow that sends notifications or newsletters.
7. Install Ollama on your local machine. Pull one mid-size model (I use llama3.1:8b). Wire it into n8n via the HTTP Request node pointing at localhost:11434.
8. Get a Claude API key from Anthropic. Add it to n8n as a credential. Use it only in workflows where output quality genuinely matters.

The whole stack can be live in a weekend. After two weeks, audit your Claude API spend in the Anthropic console. If it is above $5 for a month at your volume, look at which workflows are calling the API and whether Ollama could handle them instead.

Start with the VPS and n8n setup. That single move eliminates the largest subscription fee in most solopreneur stacks. Once automation is running locally, the free-tier infrastructure slots in around it with almost no friction.