Building a Crypto Intelligence API as a Solo Founder

I built Regime — a crypto market intelligence API that classifies markets as bull/bear/chop using 10 weighted signals — in about 4 weeks as a solo founder. Here's the architecture, costs, and what I'd do differently.

The Stack

Yes, $35/month. Everything else is free.

Data Sources (9 sources, $35/mo)

| Source | Cost | Data |

|---|---|---|

| Binance.US | Free | Candles, prices, volume, order book |

| CoinGlass | $35/mo | Funding rates, OI, liquidations |

| Hyperliquid | Free | Funding rates, OI (no auth needed) |

| CoinGecko | Free | Market cap, dominance, price fallback |

| Alternative.me | Free | Fear & Greed index |

| DeFiLlama | Free | DeFi TVL, stablecoin supply |

| FRED | Free | DXY, Fed funds rate, 10Y treasury |

| Yahoo Finance | Free | SPX, Gold, VIX |

The key insight: most financial data is free or near-free. The expensive stuff (Glassnode at $799/mo, Bloomberg at $24K/mo) gives you deeper data, but for regime classification you don't need tick-level on-chain metrics. You need macro indicators.

Architecture Decisions

SQLite over Postgres: For a single-server API with < 1000 RPM, SQLite is faster, simpler, and zero-maintenance. WAL mode handles concurrent reads. I'll migrate to Postgres when I need multi-server or > 10K RPM.

Home server over cloud: A Cloudflare Tunnel gives you HTTPS + DDoS protection + caching for free. My server is a desktop PC. Total hosting cost: my electric bill.

TypeScript monorepo: Everything in one repo with pnpm workspaces. Packages: auth, billing, macro-data, strategies, alerts. Apps: dashboard (API server), engine (pipeline), telegram-bot.

PM2 for everything: 9 candle analysis lanes, dashboard, telegram bot, daily report cron, watchdog — all managed by PM2. pm2 save && pm2 startup survives reboots.

The Regime Classifier

10 signals, each votes bullish/bearish/neutral:

``

Weighted sum > 0.2 → Bull

Weighted sum < -0.2 → Bear

Between → Chop

`

Weights were determined by grid search over 302K 4-hour candles. SMA 50/200 got 20% weight (highest standalone predictive power). Funding rates got 10% each. Everything else got 5-10%.

The trick is graceful degradation: when a data source fails, its weight redistributes proportionally. The system never breaks.

What I'd Do Differently

  • Start selling earlier. I built features for 3 weeks before making the landing page. Should have had a landing page on day 1.
  • Simpler tier structure initially. Started with feature-gating, switched to usage-limiting. Usage-limited free tiers convert 73% better.
  • Skip the Discord bot. Telegram + X + API is enough. Discord communities need 50+ active users to not feel dead.
  • Build the blog earlier. SEO compounds. Every day without indexed content is a day of lost organic traffic.

    • Current Numbers (Day 25)

    • 10 signups (5 real users)
    • 1 paying customer (Pro, $49/mo)
    • 6,730+ market snapshots collected
    • 14 SEO blog posts
    • $35/mo costs, $49/mo revenue
    • 25 days of uptime

    Not going to win any awards yet. But the data moat grows every hour, and the distribution machine is running on autopilot.

    Try It

    `bash

    curl https://getregime.com/api/v1/market/regime

    ``

    Free, no auth needed. Quickstart guide

    Building in public on X.