tape

AI-assisted stock + crypto trading terminal. Named for reading the tape — the scrolling quote strip is the first thing you see. Static frontend (GitHub Pages) + Cloudflare Worker backend holding your secrets, wired to Alpaca (one API for both US stocks and crypto, with identical paper and live endpoints) and the Anthropic API for the analyst/autopilot brain.

index.html   → the dashboard (host on GitHub Pages)
worker.js    → Cloudflare Worker (auth, Alpaca proxy, AI analysis, guardrails)
wrangler.toml→ Worker config (mode + trade caps)

1. Get accounts & keys

1. Alpaca — sign up at alpaca.markets. Every account gets a paper trading environment instantly with $100k fake money. Generate API keys from the dashboard (paper keys and live keys are separate — start with paper keys).
2. Anthropic (optional) — get an API key from console.anthropic.com if you want the AI Analyst + Autopilot. Skip it and tape runs as a clean manual trading terminal; the AI panel disables itself automatically.
3. Invent an AUTH_KEY — any long random string. This is the password your dashboard uses so nobody else can drive your worker.

2. Deploy the worker

npm i -g wrangler
wrangler login
cd neontrader
wrangler secret put ALPACA_KEY
wrangler secret put ALPACA_SECRET
wrangler secret put AUTH_KEY
wrangler secret put ANTHROPIC_API_KEY   # optional — only for AI features
wrangler deploy

The worker is named trader in wrangler.toml, so it deploys to https://trader.detlaffcameron.workers.dev — which the frontend is already pointed at.

3. Deploy the frontend

Push index.html to a GitHub Pages repo (or just open it locally — it's fully static). On first load the worker URL is pre-filled; you only enter your AUTH_KEY. Both are stored in localStorage.

4. Trade (paper first!)

• Watchlist — stocks as plain tickers (NVDA), crypto as pairs (BTC/USD).
• Order ticket — market/limit, sized in dollars or units.
• AI Analyst — pulls ~90 recent bars + snapshot + your position, sends it to Claude, gets back a strict-JSON verdict (buy/sell/hold, confidence, size, reasoning, risk). You click Execute.
• Autopilot — cycles through your watchlist on a timer and auto-executes when confidence ≥ your threshold. Only runs while the tab is open.

5. Going live (real money)

Only after you're happy with paper results:

1. Fund your Alpaca live account.
2. Generate live API keys and re-run wrangler secret put ALPACA_KEY / ALPACA_SECRET with them.
3. In wrangler.toml, set MODE = "live" and wrangler deploy.

The header badge turns red and pulses, every manual/AI order gets a confirm dialog, and autopilot demands an explicit confirmation when engaged.

Guardrails (server-side, can't be bypassed by the UI)

Var	Default	What it does
MODE	paper	Which Alpaca environment orders go to
MAX_NOTIONAL	200	Rejects any order above this many USD
MAX_TRADES_PER_DAY	20	Rejects orders past this count per UTC day (best-effort; resets if the worker isolate recycles)

Honest warnings

• AI trading is not a money printer. Claude analyzes recent price data; it has no crystal ball, and short-horizon technical signals are weak. Expect losses, especially with autopilot. Keep MAX_NOTIONAL small.
• Stock market data uses Alpaca's free IEX feed (slightly limited); crypto data is 24/7.
• Stocks only fill during US market hours; crypto trades round the clock.
• Fractional/notional stock orders must be market orders on Alpaca.
• This is a personal tool, not financial advice, and past paper performance guarantees nothing live.

Ideas for later

• Persist the daily trade counter in Workers KV so it survives isolate recycles.
• Add stop-loss brackets (order_class: "bracket") to every AI buy.
• Log every AI decision to Firestore and score its hit rate over time before ever going live.