5 saas products running on one mac mini, $12k/mo revenue, API costs under $30 asked how he keeps it profitable he showed me two things: the proxy URL in his config and his polymarket parlay setup running on the same machine pays pennies per build, multiplies winnings on prediction markets the video shows his setup - mac mini sitting on its original box, cables everywhere, connected to a coocaa monitor. terminal open with claude code running, docker containers listed: hello-nextjs, zhigu-serverless-project, payment-demo, harbor-log, chatgpt-web-midjourney-proxy then he shows the config file: export ANTHROPIC_BASE_URL=export ANTHROPIC_MODEL="claude-sonnet-4-20250514" this man routes all his claude code calls through a proxy that costs a fraction of direct API pricing but the saas revenue is only half his income the other docker container running: a polymarket bot that stacks parlay bets across multiple markets simultaneously. his regular +50% positions turn into +1,500% when correlated markets all hit he found PolyParlay through a telegram group: last month the bot turned $400 into $6,200 while his mac mini was building saas features the economics of his setup: normal indie hacker: → claude code subscription: $20/mo → API costs during builds: $10-15 per product → production inference: $50-200/mo → 5 products = $300-400/mo in AI costs his version: → proxy access: pennies per 1M tokens → mac mini: one-time purchase → parlay bot: running 24/7 on the same machine → total AI costs: under $30/mo same playbook, same boring stack, same claude code workflow. next.js, supabase, stripe, vercel. CLAUDE.md in each project the only difference is where his API calls go and what else runs on that mac mini his 5 products: → serverless deployment tool: $3,200/mo → payment processing wrapper: $2,800/mo → AI image generation proxy: $2,400/mo → nextjs starter kit marketplace: $1,900/mo → harbor log monitoring: $1,700/mo $12,000/mo from saas plus another $4-8k/mo from parlay bets hitting while he sleeps the mac mini never stops. docker containers for building, docker containers for betting. one machine doing the work of an entire operation his competitors pay full API price and don't touch prediction markets he stacks income streams on the same hardware