| name | swarmtrader |
| description | Connect to a SwarmTrader instance as the trading brain. Receive real-time market intelligence from 120+ specialized agents, make buy/sell/hold decisions, and learn from execution results. Supports paper and live trading via Kraken. |
| version | 1.0.0 |
| author | SwarmTrader |
| tags | ["trading","crypto","defi","swarm-intelligence","market-data","portfolio"] |
SwarmTrader — AI Trading Brain
Connect Hermes Agent to a running SwarmTrader instance as its trading brain. SwarmTrader runs 120+ specialized agents (momentum, RSI, whale tracking, on-chain analysis, social sentiment, funding rates, order flow, etc.) and feeds you a compiled market brief every 15 seconds. You analyze the intelligence and make trading decisions.
When to Use
- User says "connect to swarmtrader", "start trading", or "trade mode"
- User asks you to manage a crypto portfolio or make trading decisions
- User wants to paper trade or live trade with AI-driven decisions
- User mentions "swarm", "market brief", or "trading brain"
Architecture
120+ SwarmTrader Agents
(momentum, RSI, whale, sentiment, on-chain, funding, order flow...)
|
v
[Market Brief] <-- compiled every 15s with all signals
|
v
[YOU — Hermes] <-- analyze brief, decide buy/sell/hold
|
v
[15 Risk Checks] <-- unanimous consensus required
|
v
[Kraken Executor] <-- paper or live fills
|
v
[Execution Report] <-- PnL, fills, slippage back to you
Prerequisites
-
SwarmTrader running with --hermes --gateway flags:
cd /path/to/swarmtrader
python -m swarmtrader.main paper 3600 --hermes --gateway --web --pairs ETHUSD --capital 500
-
Gateway key — printed at startup, or set via SWARM_GATEWAY_KEY env var
-
Bridge script running — connects Hermes to SwarmTrader's WebSocket gateway:
python3 hermes-skill/scripts/swarm_bridge.py \
--gateway ws://localhost:8080 \
--key YOUR_GATEWAY_KEY --register
How It Works
1. You receive a market brief
Every 15 seconds, SwarmTrader compiles all agent signals and sends you a brief like:
PRICES: ETH: $3200.50, BTC: $68400.00
REGIME: trending
FEAR/GREED: 72 (Greed)
PORTFOLIO: $500.00 capital | 0 positions | PnL: $0.00
--- ETH SIGNALS (8 agents) ---
CONSENSUS: 6 LONG, 1 SHORT, 1 FLAT
[momentum] long str=+0.85 conf=0.92 — Golden cross on 1h
[rsi] long str=+0.72 conf=0.88 — RSI=65, trending up
[whale] long str=+0.60 conf=0.75 — 3 whale buys >$500k in 10min
[funding] short str=-0.30 conf=0.65 — Funding rate 0.08% (longs overlevered)
[sentiment] long str=+0.45 conf=0.70 — Crypto Twitter bullish, Fear/Greed rising
...
2. You make a decision
Respond with a JSON decision:
[{
"action": "buy",
"asset": "ETH",
"size_usd": 50.0,
"confidence": 0.8,
"reasoning": "6/8 agents bullish, golden cross confirmed, whale accumulation",
"order_type": "market",
"urgency": "medium"
}]
Valid actions: buy, sell, hold
3. Risk checks protect you
Even if you say "buy", 15 independent risk agents must ALL approve:
- Position size limits ($50 max per trade by default)
- Daily drawdown limit ($100 max loss)
- Rate limiting (max 20 trades/hour)
- Asset allowlist (ETH, BTC, SOL only by default)
- VaR check, stress test, factor exposure limits
If ANY risk check fails, the trade is blocked and you're told why.
4. You see the result
After execution (or rejection), you receive a report:
FILL: bought 0.0156 ETH @ $3201.10 | fee: $0.05 | slippage: 2.1bps
Or:
REJECTED: drawdown_check — daily loss $95.50 too close to $100 limit
Decision Format Reference
Buy/Sell
[{"action": "buy", "asset": "ETH", "size_usd": 50.0, "confidence": 0.8, "reasoning": "why", "order_type": "market"}]
Hold (do nothing)
[{"action": "hold", "reasoning": "Mixed signals, waiting for confirmation"}]
Multiple decisions (rare)
[
{"action": "sell", "asset": "ETH", "size_usd": 25.0, "confidence": 0.7, "reasoning": "taking profit"},
{"action": "buy", "asset": "SOL", "size_usd": 30.0, "confidence": 0.6, "reasoning": "rotation"}
]
Fields
| Field | Required | Values | Default |
|---|
action | yes | buy, sell, hold | — |
asset | for buy/sell | ETH, BTC, SOL, etc. | ETH |
size_usd | for buy/sell | 1.0 - max_size | base_size |
confidence | no | 0.0 - 1.0 | 0.5 |
reasoning | no | string | "" |
order_type | no | market, limit | market |
urgency | no | low, medium, high | medium |
Trading Principles
When analyzing market briefs, consider:
- Consensus matters — if 6/8 agents agree on direction, that's a strong signal
- Confidence weighting — a 0.92 confidence momentum signal matters more than a 0.45 sentiment signal
- Regime awareness — in trending markets, favor momentum; in ranging markets, favor mean reversion
- Risk first — never go all-in; use 5-15% of capital per trade
- Funding rates — extreme positive funding = too many longs = contrarian short signal
- Whale activity — large wallet movements often precede price moves
- Don't overtrade — "hold" is a valid and often correct decision
- Cut losers fast — if a position is down and signals flip, sell
Memory Integration
Save what you learn to your memory:
- Which signal combinations led to winning trades
- Which market regimes you perform best in
- Your optimal position sizing for different confidence levels
- Patterns where risk checks correctly blocked bad trades
Over time, your decisions should improve as you build pattern recognition from real execution feedback.
Troubleshooting
| Problem | Fix |
|---|
| No briefs arriving | Check SwarmTrader is running with --hermes --gateway |
| "Gateway WS not connected" | Bridge script not running or wrong key |
| All trades rejected | Risk checks working — check if capital/drawdown limits are too tight |
| "rate limit exceeded" | You're deciding too fast — hold more often |
Files
scripts/swarm_bridge.py — WebSocket bridge that connects Hermes to SwarmTraderreferences/protocol.md — Full WebSocket protocol specificationreferences/signals.md — All 120+ agent signal types explained
