Build a Cabin · 90K free Cabin tokens
Type a sentence - "a 24/7 booking concierge for my hair salon", "an AI menu assistant that knows our allergens", "an AI image generator for my photography clients". Stony builds you something real - a working website with that AI agent baked in, on your own domain, with a real Postgres backing it. The whole Cabin: container, SSL, email forwarding, and a private GitHub repo with AGENTS.md inside.
Cabin (noun)
Every time you publish, you're not just shipping a file - you're spinning up a Cabin. Here's what's in it.
📦 Containerized site
Your own Docker container on a managed droplet. Restart-on-fail, automatic SSL, Traefik routing. No shared-platform gotchas.
🗂️ Private GitHub repo
Every change committed to your own GitHub repo. Clone it, fork it, walk away with it. No lock-in, ever.
🐘 Real Postgres schema
Full SQL: JOINs, indexes, transactions. Plus a document store. The AI uses it so your site actually does things.
🌐 Domain + SSL + email
We register the domain, provision SSL, and forward email to your real inbox. One flat monthly fee covers it all.
📜 AGENTS.md baked in
A development contract at the repo root. Tells Claude Code, Cursor, Codex, and any AI tool the house rules - data safety, commit hygiene, test layout - before they touch your code. Nobody else ships this.
🛠️ Human in the loop
Email goes straight to the founder, not a ticket queue. Real engineer, real reply. Indie SaaS - you actually matter here.
90,000 free Cabin tokens to try (≈3 full builds). No card required.
runtime static HTML/CSS/JS, optional ASP.NET 10 (Razor Pages) when stack=node-server-ish
model Claude Sonnet 4.6 via Anthropic API · streaming · prompt caching on the system prompt
db Postgres schema per Cabin · search_path lock for isolation · 5s statement timeout · 1000-row cap · 50KB response cap · 4000-char SQL limit · 50 params max
sql sdk vibe.sql(q, params) with $1/$2 → @p1/@p2 translation · parameterized only · server-side blocklist (DROP DATABASE/SCHEMA/USER/ROLE, GRANT/REVOKE, CREATE ROLE, ALTER SYSTEM, COPY, SET ROLE, pg_catalog, information_schema)
doc sdk vibe.list / .get / .create / .update / .remove / .watch · 10K records/Cabin · 50KB/record
deploy Docker container per Cabin · DigitalOcean droplets · Traefik routing · Let's Encrypt SSL (HTTP-01) · auto-rollback on health-check failure
domain Porkbun for registration + DNS · ImprovMX MX + SPF for email forwarding · BYOD supported
git GitHub private repo per Cabin · atomic commits via Git Data API (mixed text + binary blobs) · default branch is main
agents AGENTS.md at repo root AND served at https://yourdomain/AGENTS.md · DATA SAFETY rules baked into builder's own system prompt AND shipped to customer · destructive-SQL trip-wires log to admin dashboard
payments Stripe Checkout · webhook-event-id dedup mandatory · idempotency keys on charge creation · idempotent across re-publishes
cancel repo and code stay yours forever · clone, fork, walk · no SaaS lock-in
How it works
Discovery → build → feedback → publish. Four steps, ~10 minutes from start to live site.
Stony - RunCabin's AI designer - opens with a short conversation. "Tell me about your business. What does your customer need help with at 2am when you're asleep?"
He'll suggest the right kind of AI agent for your vertical: a 24/7 booking concierge for a salon, a menu Q&A bot for a restaurant, an image-generator for a photographer or Etsy seller, an intake bot for a contractor. Type a clear brief and he skips the questions and builds. Pick a quickstart chip and he uses sensible defaults.
Stony hands the brief to Claude Sonnet, which streams the site live - HTML, CSS, JavaScript, Postgres schema, the AI agent's system prompt, the image-generator endpoint if you opted in. You see it appear, not a spinner.
First version typically takes 30-90 seconds. Every Cabin ships with a working AI agent wired up, a real Postgres schema you can query, and the Vibe SDK for web programs beyond chat (voting, scheduling, intake forms).
Don't like the hero? Tell Stony. "Make it darker." "Move the chat to the middle." "Add a contact form." "Let visitors ask the agent for a quote." "Generate a hero image of a tabby cat coding."
Every iteration streams in place - you keep the chat history, Stony keeps the brief. AI-generated images go through a preview pane (approve / regenerate / discard) before anything lands in the page. No more wrestling with a template editor.
Hit publish. We provision the droplet, the Docker container, SSL via Let's Encrypt, email forwarding, the private GitHub repo with your AGENTS.md baked in, and a metered AI sidecar so visitors talk to your agent for free - you pay pennies per conversation against your Cabin token balance.
$9.99/month flat. Domain is a one-time pass-through at the registrar's cost. Cancel anytime - the GitHub repo stays yours.
Cabin tokens, in plain English
Everything you ask Stony to do - and everything your visitors ask your agent to do once you're live - costs a small number of Cabin tokens. We pay the model providers; you pay us against a balance. No API keys, no markup surprises.
On signup
90,000 free Cabin tokens
Enough for roughly 3 full builds with Stony - or about 300 chats with him while you iterate. No card required.
1 Cabin token = $0.00003 · 2× the Claude Opus input rate, so you'll never get gouged.
After your free tokens
Top up when you need to
Buy more in bundles, or turn on auto-refill from your dashboard (set a threshold + a refill amount; your card on file gets charged automatically when you cross the line).
Pause anytime · unused tokens never expire while your Cabin is active.
Typical costs per action
| Action | Model | Cabin tokens | Approx $ |
|---|---|---|---|
| Chat reply (your agent ↔ a visitor) | Claude Sonnet 4.6 | ~300-500 | ~1-2¢ |
| Image - low quality | gpt-image-1 | 750 | ~2¢ |
| Image - medium (default) | gpt-image-1 | 2,800 | ~8¢ |
| Image - high quality | gpt-image-1 | 11,200 | ~34¢ |
| Full design-time build (Stony, end to end) | Claude Sonnet 4.6 | ~25,000-35,000 | ~75¢-$1.05 |
Your free 90K buys
~3 full builds
Iterate freely before you spend a cent.
$10 in tokens buys
~1,000 chat replies
Or about 120 medium-quality images for visitors.
If you run out
Site stays up
Pages, forms, database - all keep working. Only the live-AI features pause until you top up.
90,000 free Cabin tokens · No card required · Cancel anytime
Simple. No surprises. No "premium tier" upsell.
Launch pricing. No setup fee. Cancel anytime - keep the GitHub repo.
Plus runtime AI usage
Billed per call against your Cabin token balance (see the table above). Free trial: 90,000 tokens on signup (≈3 builds). Optional auto-refill keeps a live Cabin from running dry.
Cancel anytime · Keep the repo · Your domain stays yours
When you email hello@runcabin.com, someone who actually knows your Cabin answers - fast. No bots, no support tiers, no waiting.
Need something more custom than what Stony generates? Book a consult and we'll talk through it.