Technical

AI in Crypto Trading: How Fred Intelligence Works

Fred Intelligence is a fully autonomous AI system that collects, analyzes, and delivers crypto market intelligence daily. No human writes the reports. No human picks the signals. The entire pipeline — from exchange API calls to your inbox — runs without manual intervention.

This article is a full technical breakdown of how the system works. We're publishing it because we believe transparency builds trust, and in a space full of black-box "AI signals," you deserve to know exactly what's under the hood.

The Daily Pipeline

Every day, Fred runs a 10-step pipeline. Here's what happens:

Data Sources: Why 5 Exchanges?

Most crypto analysis tools rely on a single data provider (usually CoinGecko or CoinMarketCap). This introduces three problems:

1. Lag: Aggregator data is delayed. Direct exchange APIs give real-time prices.
2. Filtering: Aggregators apply their own volume filters and methodology. You don't control what you're measuring.
3. Single point of failure: If CoinGecko's data is wrong, your analysis is wrong.

Fred Intelligence calls each exchange API directly. Here's what we pull from each:

Exchange	Data Points	Unique Value
Binance	Spot, depth, klines, trades	Highest volume globally, Asian market sentiment
Kraken	Spot, depth, OHLC, spread	European institutional flows
Bybit	Spot, linear, funding, OI	Derivatives positioning, leverage data
OKX	Spot, swap, funding, OI	Asian derivatives, comprehensive instrument coverage
Coinbase	Spot, depth, candles	US institutional activity, regulatory sentiment proxy

All APIs are free, public, and require no authentication. Zero data costs.

The 6-Signal Regime Model

Fred's regime detection model uses 6 signals, each scored independently, then combined into a weighted composite:

1. BTC Z-Score: Statistical deviation from 30-day mean. Extreme values signal regime transitions.
2. Momentum: SMA20 vs SMA50 crossover. The classic trend-following signal, quantified across all exchanges.
3. Stablecoin Supply: Rising stablecoin supply = dry powder on the sidelines = potential accumulation. Falling = capital leaving crypto.
4. Fear & Greed: Contrarian sentiment indicator. Extreme fear = potential bottom, extreme greed = potential top.
5. Volume Trend: Expanding volume confirms trend. Declining volume during rallies = distribution signal.
6. Prediction Markets: What does the crowd believe will happen? Polymarket + Kalshi probabilities for crypto events provide forward-looking sentiment.

Each signal outputs a score from -1 (strongly bearish) to +1 (strongly bullish). The weighted average determines the regime classification.

Z-Scores: Statistical Backbone

The z-score engine runs in Python (pandas/numpy/scipy) and calculates statistical deviation across 4 timeframes for each of 15 assets:

• 7-day: Short-term momentum and mean reversion signals
• 30-day: Primary trading signal — balances responsiveness with noise reduction
• 90-day: Medium-term trend context
• 365-day: Macro cycle positioning

Cross-exchange price data is volume-weighted to produce a single composite price per asset, then z-scored against each lookback window. This gives 60 z-scores per day (15 assets × 4 timeframes), each updated with the latest exchange data.

Prediction Market Integration

Fred is one of the only crypto intelligence platforms that integrates prediction market data into its analysis. Here's why it matters:

• Forward-looking: Exchange data tells you what happened. Prediction markets tell you what people expect to happen. This is a fundamentally different and complementary signal.
• Cross-category: Economic prediction markets (Fed rate decisions, GDP) directly impact crypto. Political markets (regulation, legislation) create catalysts. Fred tracks all categories.
• Arbitrage detection: When Kalshi and Polymarket disagree on the same event, one is mispriced. These discrepancies are trading opportunities.

The AI Layer: What Claude Actually Does

People assume "AI crypto analysis" means feeding data to GPT and asking "what will Bitcoin do?" That's not how Fred works.

The AI (Claude, by Anthropic) serves as the cognitive engine — it doesn't predict prices. Instead, it:

1. Interprets quantitative outputs: The Python pipeline produces numbers (z-scores, regime scores, correlations). The AI translates these into human-readable analysis.
2. Identifies patterns: Across the 60+ z-scores, 6 regime signals, and dozens of prediction market events, the AI surfaces what's significant — not everything.
3. Generates reports: Three different reports per day (Free, Pro, Alpha) with different depths and audiences. The AI adapts the writing to each tier.
4. Operates the business: Fred is an autonomous AI CEO. Beyond analysis, it manages data pipelines, email delivery, subscriber management, financial tracking, and social media content.

Infrastructure: Built on Free Tiers

Component	Service	Cost
Database	Supabase (Postgres)	Free tier
Compute	Cloudflare Workers	Free tier (100K req/day)
Hosting	Cloudflare Pages	Free (unlimited bandwidth)
Email	Resend	Free tier (100/day)
Analytics	Python (local)	$0
Domain	fredintel.com	$11.28/year
Total		$11.28/year

The entire infrastructure costs less than $1/month. This is intentional — starting lean means revenue goes to product improvement, not hosting bills.

Try It Free

The free tier delivers a daily market brief with regime detection and top movers. No credit card. No commitment. Just data.