DeployHow much does it cost to run rakazo? A real-numbers breakdown
Rakazo itself is free (Apache-2.0 open source). Running it is not quite free: you pay for a server and for the model your bots talk to. Below is what a real, always-on setup costs — anchored on our own test server's actual bill, with provider list prices for the rest. Checked October 2026; prices move, so treat the totals as a budget, not a quote.
1. The server: about €5–13/month
Our test instance runs the full stack (app, database, and the bots' sandbox computers) on a Hetzner CAX21 — 4 vCPU / 8 GB RAM / 80 GB, Nuremberg. Our actual charge: €13.08/month including VAT, billed hourly. That is the comfortable size we recommend: the stack idles at roughly 1.5 GB, and 8 GB leaves room for several bots working at once.
- Smaller plans (2 vCPU / 4 GB class) list from roughly €4–5/month before VAT and can work for a single light bot — we have not production-tested that size, so we do not anchor budgets on it.
- List prices are shown without VAT on Hetzner's site and vary by region; Hetzner also adjusted cloud prices during 2026. Check hetzner.com for the figure that applies to you.
- Optional: automatic backups add about 20% to the server price (≈ €2.60/month on a CAX21).
2. The model: from €0 extra to a few dollars
You bring your own model access, and there are two routes:
Route A — a subscription you already have. Connect a supported Claude or ChatGPT subscription and the marginal cost is €0: your bots draw on the plan you are already paying for. (Plan limits apply, and heavy bot use can hit them.)
Route B — an API key, pay per token. Worked examples at OpenAI list prices, October 2026 (see OpenAI's pricing page for current figures):
| Model (list price per 1M tokens) | Light month: 2M in / 0.5M out | Heavy month: 20M in / 5M out |
|---|---|---|
| GPT-6 Luna — $0.10 in / $0.50 out | $0.45 | $4.50 |
| GPT-5 mini — $0.25 in / $2.00 out | $1.50 | $15.00 |
| GPT-6.1 Sol — $2.00 in / $10.00 out | $9.00 | $90.00 |
Token volumes are illustrations, not measurements: a "light month" is a personal team doing daily briefs, drafts and file tasks; a "heavy month" is near-constant agent work. Your provider dashboard is the only source that knows your real usage. Frontier models cost 20–200× the mini-class ones for the same tokens — most people run cheap models for routine bots and a strong model only where it matters.
3. The small print
- Domain (optional): a custom domain for your instance, about $10/year at cost-price registrars (ours: deploybutler.com, $10.46/year). A free sslip.io address works if you skip this.
- Email for password recovery (optional): an SMTP provider's free tier is enough at this scale; without it, recovery means a database procedure (we wrote the forgot-password guide for exactly that case).
- Your time: our timed install took about 25 minutes on a prepared server — plus the pitfalls (email verification, signup lockdown, backups) that our install guide exists to help you dodge.
The total
| Self-hosted rakazo | Hosted route (Grok Bot via X) | |
|---|---|---|
| Monthly running cost | ≈ €13–18 (server + light model use) | $30–300 subscription, before any model extras |
| One-time setup | Your evening, or $49–149 with us | — |
| Where the data lives | Your server | The platform's cloud |
That is the whole economic argument for self-hosting: roughly half the entry price of the hosted route, and the expensive line item (frontier-model tokens) is one you control per bot.
Our server charge (€13.08/month incl. VAT) is from our October 2026 Hetzner billing. Model prices are OpenAI list prices checked October 2026. Grok Bot access pricing follows X Premium / SuperGrok tiers as published in October 2026. This page is a budget guide, not a quote.