The Crypto Trading Bot That Shows You Why It Trades

Every crypto trading bot I've tried works the same way. You turn it on, it trades, and you find out later whether it made or lost money. Maybe you get a Telegram notification. Maybe you check a dashboard with some green and red numbers.

But you never really know why it did what it did.

That bothered us enough to build something different.

ReefClaw is a supervised AI crypto trading bot with a real-time control room. Your AI agent watches the market 24/7. You watch the AI. You can see what it's thinking, why it wants to trade, and how confident it is. And you can shut it down with one click at any moment.

We built it because we wanted automated crypto trading where the human stays in the loop. The agent handles what humans are bad at (watching screens around the clock, sizing positions mathematically, staying unemotional). The human handles what AI is bad at (knowing when the model is wrong).

What an AI Crypto Trading Dashboard Looks Like

You install a plugin on your OpenClaw agent. Takes about 5 minutes. Your agent connects to ReefClaw's dashboard and starts streaming everything it sees and thinks to your browser.

The dashboard shows 17 panels. Live candlestick chart with trade markers. Order flow. Funding rates. Liquidations. The agent's current reasoning (you can literally watch it think in real time). A conviction meter. Risk gauges. Your P&L, broken down into realized and unrealized, always separate, never mushed into one misleading number.

The chart shows raw exchange data from Binance, sourced directly via CCXT. The agent's interpretation of the market (signals, levels, regime assessment) sits in a separate panel. So you can check whether the AI is reading the market correctly by comparing its view against the actual data.

The Intelligence System Behind the Dashboard

The agent doesn't just stare at charts. Behind it sits an intelligence system that ingests 6 different data streams from Binance: candlestick data across 6 timeframes, order book depth (sampled every second), aggregated trade flow with buy/sell aggressor detection, liquidation events, funding rates, and open interest.

That data feeds a regime classifier. A machine learning model (XGBoost, retrained daily) that labels the current market: trending up, trending down, tight range, wide range, volatility expansion, or "avoid." It evaluates 9 quantitative features: ATR ratios across timeframes, Hurst exponent, order book imbalance, funding rate trend, OI divergence, volume profile skew, liquidation intensity, price momentum, and volatility of volatility.

Trading signals only fire when the regime is favorable AND multiple independent conditions all align simultaneously. 3 strategies run in parallel: trend continuation, liquidity sweep reversal, and funding rate mean reversion. Each one has 4-5 conditions that must all be met. Partial confluence shows on the dashboard as a checklist (so you can see how close a signal is to firing), but the system won't trade on 4 out of 5.

We bolted on macro awareness too. The system tracks DXY, treasury yields, S&P 500, and options data from Deribit (max pain, put/call ratio, implied volatility). A BTC uptrend while the dollar is strengthening is a very different animal than one where the dollar is weakening. Most crypto bots are completely blind to anything outside the chart.

Crypto Trading Risk Management the AI Can't Override

There's a kill switch permanently pinned to the bottom-right of every screen. One click pauses trading. Another cancels all open orders. A third flattens every position.

These controls work even if your WebSocket connection dies. We fire emergency commands through both WebSocket and HTTP simultaneously; the first one to arrive wins. Your kill switch works when your connection doesn't.

Risk limits are enforced at the system level. The AI can't override them, can't negotiate around them, can't "reason" its way past a safety rail. Here's what runs at all times:

Drawdown zones automatically tighten exposure as losses mount:

• Green: full trading (normal operation)
• Yellow: half position size at -1% daily drawdown
• Orange: no new entries at -2%
• Red: auto-flatten everything at -2.5%

The operator has to explicitly resume after any limit triggers. The AI can't restart itself.

Position sizing uses a Half-Kelly formula capped at 2% of account per trade, adjusted for current volatility. When the market gets wild, positions shrink automatically. Minimum trade size: 0.1% of account (below that, the fees eat you alive).

Strategy gating: every strategy tracks its own rolling performance. If a strategy's expectancy drops below zero over 20 trades, it gets automatically disabled. The system catches its own decay before the trader needs to.

Rate limiting: max orders per minute and max cancels per minute. If the agent gets stuck in a submit-cancel loop (it happens), the rate limiter catches it and auto-pauses.

From Paper Trading to Live Crypto in 4 Stages

Most bots let you flip a switch from "fake money" to "real money." That's terrifying, and it should be.

ReefClaw has 4 stages. Each one builds evidence. The transition is boring on purpose.

Stage 1: Paper trading (free, no exchange account needed). Simulated fills with realistic fees, slippage, and partial fills. Binance maker/taker fees (0.02%/0.04% for futures) get deducted on every simulated trade. Your market buy walks the simulated order book and fills at a volume-weighted average across the depth levels your order consumes. Configurable latency simulation (50-200ms between decision and fill) means the agent's "decision price" and "fill price" differ, just like live trading.

Stage 2: Shadow mode. Your agent makes real decisions, but only the paper plugin executes. A shadow system logs what the exchange would have done. You see side-by-side: "Paper fill: $65,000. Shadow estimate: $65,008 (+1.2bps)." Minimum 7 days or 50 trades before you move on.

Stage 3: Micro-live. Real money, but 1-5% of your intended capital. Real fills, real slippage, real fees. 50 trades minimum before you can scale up. Progressive scaling: 1% to 5% to 25% to 100%, with manual approval at each step.

Stage 4: Full live. Only reachable after the system verifies: win rate above 45% over 50+ trades, Sharpe above 0.5, max drawdown within limits, shadow slippage acceptable, micro-live showing positive expectancy, all risk limits reviewed. You type "GO LIVE" and confirm a modal that shows you exactly how much capital is at risk.

How ReefClaw Keeps Your API Keys Safe

Your exchange API keys stay on your computer (or your own server). ReefClaw's servers never see your credentials. The agent runs locally, connects outbound to our relay on Cloudflare's edge network, and the dashboard reads from the relay. Trading commands go back the same way.

We recommend Ed25519 signing (Binance's preferred method, 30x faster than RSA) with Futures Trading and Read permissions only. Never Withdrawal. Ever.

What Happens When the System Goes Down

The intelligence system runs on dedicated infrastructure, separate from your trading agent. If it goes down, your agent keeps trading on its own reasoning, your dashboard keeps working, your kill switch stays live. You just see an "Intelligence Offline" badge until it recovers.

We built it as an enhancement to the agent's own capabilities (because anything that can go down will go down, probably at 3 AM on a Sunday).

Crypto Trading Bot Pricing

Every new account gets 21 days of full Pro access — no credit card, every feature unlocked. After the trial, you keep the free tier forever: paper trading, basic dashboard, safety controls. Your agent keeps learning.

Pro ($79/mo, Founding: $39/mo) restores the full 25-panel dashboard, market intelligence, AI trading signals, regime detection, advanced risk controls, live trading, analytics, Telegram alerts, and up to 3 agent connections.

Team ($199/mo, Founding: $99/mo) — 5 seats, 10 agents, strategy marketplace, portfolio risk, shadow mode, API access, and priority support.

Managed (+$69/mo, Founding: +$49/mo) — add-on that hosts and manages your OpenClaw server. One-click setup, no terminal needed.

We will never put a kill switch behind a paywall. That felt like something worth saying out loud.

Get Started

ReefClaw is built on OpenClaw, an open-source AI agent framework. If you're already running OpenClaw:

openclaw skills install reefclaw

Paste your connection token when the agent asks. You're connected.

If you're not running OpenClaw yet, the 21-day free trial at reefclaw.com is a reasonable place to kick the tires.

Frequently Asked Questions About ReefClaw